Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1394

Last change on this file since 1394 was 1379, checked in by raasch, 11 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 50.3 KB

Rev	Line
[1]	1	#!/bin/ksh
[1090]	2
	3	# subjob - script for automatic generation and submission of batch-job files
	4	# for various batch queuing systems
	5
[1046]	6	#--------------------------------------------------------------------------------#
	7	# This file is part of PALM.
	8	#
	9	# PALM is free software: you can redistribute it and/or modify it under the terms
	10	# of the GNU General Public License as published by the Free Software Foundation,
	11	# either version 3 of the License, or (at your option) any later version.
	12	#
	13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
	14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
	15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
	16	#
	17	# You should have received a copy of the GNU General Public License along with
	18	# PALM. If not, see <http://www.gnu.org/licenses/>.
	19	#
[1310]	20	# Copyright 1997-2014 Leibniz Universitaet Hannover
[1046]	21	#--------------------------------------------------------------------------------#
	22	#
	23	# Current revisions:
[1090]	24	# ------------------
[1351]	25	#
[1379]	26	#
[1046]	27	# Former revisions:
	28	# -----------------
[169]	29	# $Id: subjob 1379 2014-04-28 06:05:51Z keck $
[1200]	30	#
[1379]	31	# 1378 2014-04-28 06:04:58Z raasch
	32	# -et option added for lctit
	33	#
[1351]	34	# 1350 2014-04-04 13:01:30Z maronga
	35	# location of qsub updated for lcxe6
	36	#
[1290]	37	# 1289 2014-03-04 07:12:34Z raasch
	38	# German comments translated to English
	39	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
	40	#
[1280]	41	# 1279 2014-01-28 12:10:14Z raasch
	42	# node calculation modified due to changes in mrun (tasks_per_node must not be
	43	# an integral divisor of numprocs any more)
	44	#
[1275]	45	# 1274 2014-01-09 13:14:54Z heinze
	46	# adjustments for lccrayh
	47	#
[1267]	48	# 1266 2013-12-11 12:07:34Z heinze
	49	# further adjustments for lccrayb (use msub instead of qsub)
	50	#
[1265]	51	# 1264 2013-12-09 12:46:09Z fricke
	52	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
	53	#
[1263]	54	# 1262 2013-12-09 10:57:20Z fricke
	55	# further adjustments for lccrayb
	56	#
[1261]	57	# 1260 2013-12-04 12:48:04Z raasch
	58	# jaboticaba admitted
	59	#
[1256]	60	# 1255 2013-11-07 14:43:35Z raasch
	61	# further adjustments for lccrayb
	62	#
[1225]	63	# 1224 2013-09-16 07:27:23Z raasch
	64	# first adjustments for lccrayb
	65	#
[1203]	66	# 1202 2013-07-10 16:22:07Z witha
	67	# adjustments for Forwind cluster (lcflow)
	68	#
[1200]	69	# 1199 2013-07-05 14:52:22Z raasch
	70	# adjustments for CSC Helsinki (lccrayf)
	71	#
[1185]	72	# use of cluster/express queue enabled (ibmh)
	73	# vinessa added (imuk)
[1047]	74	#
[1104]	75	# 1103 2013-02-20 02:15:53Z raasch
	76	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
	77	# instead typeset -L/R),
	78	# further adjustments for lckyuh
	79	#
[1100]	80	# 2013-02-10 01:47:43Z raasch
	81	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
	82	# and for Forwind cluster (lcflow)
	83	#
[1096]	84	# 1094 2013-02-03 01:52:12Z raasch
	85	# new option -P for explicit setting of ssh/scp port,
	86	# decalpha parts (yonsei) removed
	87	#
[1091]	88	# 2013-02-02 07:06:13Z raasch
[1099]	89	# adjustments for Kyushu-University computing center (lckyut - tatara)
[1091]	90	# old changelog messages removed
	91	#
[1047]	92	# 1046 2012-11-09 14:38:45Z maronga
	93	# code put under GPL (PALM 3.9)
	94	#
[1090]	95	# 08/07/94 - Siggi - first version finished
	96	# 29/06/94 - Siggi - script development started
	97	#--------------------------------------------------------------------------------#
	98	# subjob - script for automatic generation and submission of batch-job files
	99	# for various batch queuing systems
	100	#--------------------------------------------------------------------------------#
[1]	101
	102
[1289]	103	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
[352]	104	delete_dayfile=false
[799]	105	email_notification=none
[122]	106	group_number=none
[1]	107	locat=normal
	108	no_default_queue=none
	109	no_submit=false
	110	job_catalog="~/job_queue"
	111	job_name=none
	112	local_user=$LOGNAME
	113	node_usage=shared
[475]	114	numprocs=0
[1]	115	punkte="..........................................................."
	116	submcom=qsub
	117	queue=default
	118	remote_host=none
	119	remote_user=""
	120	verify=true
	121
	122	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	123	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	124
	125
	126
[1289]	127	# ERROR HANDLING
	128	# IN CASE OF EXIT:
[1]	129	trap 'if [[ $locat != normal ]]
	130	then
	131	case $locat in
	132	(option) printf "\n --> available optios can be displayed"
	133	printf " by typing:"
	134	printf "\n \"subjob ?\" \n";;
	135	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	136	(*) printf "\n +++ unknown error"
	137	printf "\n please inform S. Raasch!\n"
	138	esac
	139	[[ -f $job_to_send ]] && rm $job_to_send
	140	printf "\n\n+++ SUBJOB killed \n\n"
	141	fi' exit
	142
	143
[1289]	144	# IN CASE OF TERMINAL-BREAK:
[1]	145	trap '[[ -f $job_to_send ]] && rm $job_to_send
	146	printf "\n\n+++ SUBJOB killed \n\n"
	147	exit
	148	' 2
	149
	150
	151
	152
[1289]	153	# DETERMINE NAME OF LOCAL HOST
[1]	154	local_host=$(hostname)
	155
	156
	157
[1289]	158	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
	159	# IS ADMITTED AT ALL)
	160	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
	161	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
[1]	162	case $local_host in
[1204]	163	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
[1]	164	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[1235]	165	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
[102]	166	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	167	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[1103]	168	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
[749]	169	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[1255]	170	(blogin\|bxc) local_addres=130.73.233.1; local_host=lccrayb;;
[1274]	171	(hlogin\|hxc) local_addres=130.75.4.1; local_host=lccrayh;;
[12]	172	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[1174]	173	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
[372]	174	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[693]	175	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
[1197]	176	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
[1]	177	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	178	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[1099]	179	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
	180	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
[367]	181	(node*) local_addres=165.132.26.61 local_host=lck;;
[717]	182	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
[693]	183	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
[101]	184	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[5]	185	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	186	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[1099]	187	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
[892]	188	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
[440]	189	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
[749]	190	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
[5]	191	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
[1260]	192	(jaboticaba) local_addres=150.163.25.181; local_host=lcbr;;
[407]	193	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
[5]	194	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
[1040]	195	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
[5]	196	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[1099]	197	(meller) local_addres=134.106.74.155; local_host=lcfor;;
[437]	198	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
[892]	199	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
[1]	200	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	201	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	202	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	203	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[1099]	204	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
[789]	205	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
[5]	206	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[622]	207	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
[1204]	208	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
[1197]	209	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
[799]	210	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
[1090]	211	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
[635]	212	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
[1021]	213	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
[1184]	214	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
[1]	215	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
[622]	216	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
[1]	217	(*) printf "\n +++ \"$local_host\" unknown";
[1255]	218	printf "\n please contact the PALM group at IMUK";
[1]	219	locat=parameter; exit;;
	220	esac
	221
	222
	223
[1289]	224	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
[1]	225	remote_host=$local_host
	226
	227
	228
	229
[1289]	230	# READ THE SHELLSCRIPT-OPTIONS
[1094]	231	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
[1]	232	do
	233	case $option in
	234	(c) job_catalog=$OPTARG;;
	235	(d) delete_dayfile=true;;
	236	(D) no_submit=true;;
[352]	237	(e) email_notification=$OPTARG;;
[125]	238	(g) group_number=$OPTARG;;
[1]	239	(h) remote_host=$OPTARG;;
	240	(m) memory=$OPTARG;;
	241	(n) job_name=$OPTARG;;
	242	(N) node_usage=$OPTARG;;
	243	(O) threads_per_task=$OPTARG;;
[1094]	244	(P) scp_port=$OPTARG;;
[1]	245	(q) no_default_queue=$OPTARG;;
	246	(t) cputime=$OPTARG;;
	247	(T) tasks_per_node=$OPTARG;;
	248	(u) remote_user=$OPTARG;;
	249	(v) verify=false;;
	250	(X) numprocs=$OPTARG;;
	251	(\?) printf "\n +++ Option $OPTARG unknown \n";
	252	locat=option; exit;;
	253	esac
	254	done
	255
	256
[1289]	257	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
[1103]	258	(( to_shift = $OPTIND - 1 ))
	259	shift $to_shift; file_to_send=$1
[1]	260
	261
[1289]	262	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
[1]	263	if [ "$1" = "?" ]
	264	then
	265	(printf "\n *** subjob can be called as follows:\n"
	266	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	267	printf "\n Description of available options:\n"
	268	printf "\n Option Description Default-Value"
	269	printf "\n -c job-input- and output-catalog ~/job_queue"
	270	printf "\n -d no job-protocol will be created ---"
	271	printf "\n -D only the job-file will be created ---"
	272	printf "\n -h execution host, available hosts: $remote_host"
[1289]	273	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
	274	printf "\n lckiaps, lctit, nech"
[1]	275	printf "\n -m memory demand per process in MByte ---"
	276	printf "\n -n jobname <jobdatei>"
	277	printf "\n -O threads per task (for OpenMP usage) 1"
[1094]	278	printf "\n -P ssh/scp port default port"
[1]	279	printf "\n -q job-queue to be used default"
	280	printf "\n -t allowed cpu-time in seconds ---"
	281	printf "\n -T tasks per node (on parallel hosts) ---"
	282	printf "\n -u username on execution host from .netrc"
	283	printf "\n -v no prompt for confirmation ---"
	284	printf "\n -X # of processors (on parallel hosts) 1"
	285	printf "\n "
	286	printf "\n The only possible positional parameter is <jobfile>:"
	287	printf "\n The complete NQS-job must be provided here."
	288	printf "\n <jobfile>=? creates this outline\n\n") \| more
	289	exit
	290	fi
	291
	292
	293
[1289]	294	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
[1]	295	if [[ "$file_to_send" = "" ]]
	296	then
	297	printf "\n +++ job-file missing"
	298	locat=parameter; exit
	299	else
	300	if [[ -f $file_to_send ]]
	301	then
	302	true
	303	else
	304	printf "\n +++ job-file: "
	305	printf "\n $file_to_send"
	306	printf "\n does not exist"
	307	locat=parameter; exit
	308	fi
	309	fi
	310
	311
	312
[1289]	313	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
	314	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
[1]	315	if [[ $job_name = none ]]
	316	then
	317	job_name=$file_to_send
	318	fi
	319	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	320	then
	321	printf "\n +++ job-file name: "
	322	printf "\n $job_name"
	323	printf "\n must not contain \"/\"-characters"
	324	locat=parameter; exit
	325	fi
	326
	327
	328
	329
[1289]	330	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
	331	# OR IF NO HOST HAS BEEN GIVEN
[1]	332	if [[ $remote_host = none ]]
	333	then
	334	printf "\n +++ host missing"
	335	locat=option; exit
	336	else
	337	case $remote_host in
[251]	338	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1184]	339	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[693]	340	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[622]	341	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
[251]	342	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[1262]	343	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/moab/7.2.6/bin/msub;;
[1274]	344	(lccrayh) queue=mpp1testq; remote_addres=130.75.4.1; submcom=/opt/moab/7.2.6/bin/msub;;
[1197]	345	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
[1099]	346	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
[440]	347	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
[367]	348	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[1040]	349	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
[717]	350	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
[1099]	351	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
[1090]	352	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
[1021]	353	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
[635]	354	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
[1350]	355	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
[437]	356	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
[693]	357	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
[251]	358	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	359	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	360	locat=parameter; exit;;
[1]	361	esac
	362	fi
	363
	364
[1289]	365	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
[1]	366	if [[ $no_default_queue != none ]]
	367	then
	368	error=false
	369	ndq=$no_default_queue
	370	case $remote_host in
	371	(ibm) case $ndq in
	372	(p690_express\|p690_standard\|p690_long) error=false;;
	373	(*) error=true;;
	374	esac;;
	375	(ibmh) case $ndq in
[1184]	376	(cluster\|express) error=false;;
[1]	377	(*) error=true;;
	378	esac;;
[693]	379	(ibmkisti) case $ndq in
	380	(class.32plus\|class.1-2\|class.2-32) error=false;;
	381	(*) error=true;;
	382	esac;;
[622]	383	(ibmku) case $ndq in
	384	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	385	(*) error=true;;
	386	esac;;
[1]	387	(ibms) case $ndq in
	388	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	389	(*) error=true;;
	390	esac;;
[1224]	391	(lccrayb) case $ndq in
[1274]	392	(mpp1q\|testq\|specialm1q) error=false;;
[1224]	393	(*) error=true;;
	394	esac;;
[1274]	395	(lccrayh) case $ndq in
	396	(mpp1q\|mpp1testq\|specialm1q\|dataq) error=false;;
	397	(*) error=true;;
	398	esac;;
[1197]	399	(lccrayf) case $ndq in
	400	(usup\|test*\|small\|large) error=false;;
	401	(*) error=true;;
	402	esac;;
[1099]	403	(lcflow) case $ndq in
	404	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
	405	(*) error=true;;
	406	esac;;
[1040]	407	(lckiaps) case $ndq in
	408	(express\|normal) error=false;;
	409	(*) error=true;;
	410	esac;;
[440]	411	(lckyoto) case $ndq in
	412	(eh\|ph) error=false;;
	413	(*) error=true;;
	414	esac;;
[1099]	415	(lckyuh) case $ndq in
	416	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
	417	(*) error=true;;
	418	esac;;
[1090]	419	(lckyut) case $ndq in
	420	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
	421	(*) error=true;;
	422	esac;;
[1]	423	(lctit) case $ndq in
[635]	424	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	425	(*) error=true;;
	426	esac;;
	427	(t3eb) case $ndq in
	428	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	429	(*) error=true;;
	430	esac;;
	431	(t3eh) case $ndq in
	432	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	433	(*) error=true;;
	434	esac;;
	435	(t3ej2\|t3ej5) case $ndq in
	436	(low\|normal\|high) error=false;;
	437	(*) error=true;;
	438	esac;;
	439	(t3es) case $ndq in
	440	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	441	(*) error=true;;
	442	esac;;
	443	esac
	444	if [[ $error = true ]]
	445	then
	446	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	447	locat=parameter; exit
	448	else
	449	queue=$no_default_queue
	450	fi
	451	fi
	452
	453
	454
[1289]	455	# CHECK THE CPU-TIME
	456	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
[1]	457	done=false
	458	while [[ $done = false ]]
	459	do
	460	if (( $cputime <= 0 ))
	461	then
	462	printf "\n +++ wrong cpu-time or cpu-time missing"
	463	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	464	printf "\n >>> "
	465	read cputime 1>/dev/null 2>&1
	466	else
	467	done=true
	468	fi
	469	done
	470	if [[ $remote_host = nech ]]
	471	then
	472	if (( tasks_per_node != 0 ))
	473	then
	474	(( cputime = cputime * tasks_per_node ))
	475	elif [[ $numprocs != 0 ]]
	476	then
	477	(( cputime = cputime * numprocs ))
	478	fi
	479	fi
	480	(( stunden = cputime / 3600 ))
	481	(( resttime = cputime - stunden * 3600 ))
	482	(( minuten = resttime / 60 ))
	483	(( sekunden = resttime - minuten * 60 ))
	484	timestring=${stunden}:${minuten}:${sekunden}
	485
	486
	487
[1289]	488	# CHECK THE MEMORY DEMAND
[1]	489	done=false
	490	while [[ $done = false ]]
	491	do
	492	if (( memory <= 0 ))
	493	then
	494	printf "\n +++ wrong memory demand or memory demand missing"
	495	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	496	printf "\n >>> "
	497	read memory 1>/dev/null 2>&1
	498	else
	499	done=true
	500	fi
	501	done
	502
[1289]	503	if [[ $remote_host = nech ]]
[1]	504	then
	505	if (( tasks_per_node != 0 ))
	506	then
	507	(( Memory = memory * tasks_per_node / 1000 ))
	508	elif [[ $numprocs != 0 ]]
	509	then
	510	(( Memory = memory * numprocs / 1000 ))
	511	else
	512	(( Memory = memory / 1000 ))
	513	fi
[635]	514	elif [[ $remote_host = lctit ]]
	515	then
	516	(( Memory = memory * tasks_per_node / 1000 ))
[1]	517	fi
	518
	519
[1289]	520	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
[1]	521	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	522	then
	523	(( memory = memory * threads_per_task ))
	524	fi
	525
	526
[1289]	527	# CALCULATE NUMBER OF REQUIRED NODES
[1]	528	if (( tasks_per_node != 0 ))
	529	then
[1279]	530	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
[1]	531	fi
	532
[1094]	533
[1289]	534	# CALCULATE NUMBER OF PROCESSES PER NODE
[622]	535	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	536
[1094]	537
[1289]	538	# CALCULATE NUMBER OF MPI TASKS
[696]	539	(( mpi_tasks = numprocs / threads_per_task ))
[1]	540
[696]	541
[1289]	542	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
[1094]	543	if [[ "$scp_port" != "" ]]
	544	then
	545	PORTOPT="-P $scp_port"
	546	SSH_PORTOPT="-p $scp_port"
	547	fi
	548
	549
[1289]	550	# HEADER-OUTPUT
[1]	551	if [[ $verify = true ]]
	552	then
	553	printf "\n\n"
	554	printf "#--------------------------------------------------------------# \n"
	555	spalte1=SUBJOB;spalte2=$(date)
[1103]	556	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
[1]	557	printf "\| \| \n"
	558	printf "\| values of parameters/options: \| \n"
[1103]	559	spalte1=$(echo local_host$punkte \| cut -c-20)
	560	spalte2=$punkte$local_host
	561	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	562	spalte1=$(echo remote_host$punkte \| cut -c-20)
	563	spalte2=$punkte$remote_host
	564	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	565	spalte1=$(echo queue$punkte \| cut -c-20)
	566	spalte2=$punkte$queue
	567	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	568	spalte1=$(echo memory$punkte \| cut -c-20)
	569	spalte2="$punkte$memory mb"
	570	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	571	spalte1=$(echo cputime$punkte \| cut -c-20)
	572	spalte2="$punkte$cputime sec"
	573	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
	574	spalte1=$(echo job_name$punkte \| cut -c-20)
	575	spalte2="$punkte$job_name"
	576	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
[1]	577	printf "#--------------------------------------------------------------# \n\n"
	578
	579
[1289]	580	# QUERY CHECK
[1]	581	antwort="dummy"
	582	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	583	do
	584	read antwort?" >>> continue (y/n) ? "
	585	done
	586	if [[ $antwort = n \|\| $antwort = N ]]
	587	then
	588	locat=verify; exit
	589	fi
	590	printf "\n"
	591	fi
	592
	593
	594
[1289]	595	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
	596	identifier=$RANDOM
	597	job_on_remhost=${job_name}_${identifier}_$local_host
	598	job_to_send=job_to_send_$identifier
[1]	599	if [[ $delete_dayfile = false ]]
	600	then
[1289]	601	remote_dayfile=${local_host}_${job_name}_result_$identifier
[1]	602	local_dayfile=${remote_host}_${job_name}
	603	else
	604	remote_dayfile=/dev/null
	605	fi
	606
	607
	608
[1289]	609	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
[1]	610	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	611	then
	612
[1289]	613	# GENERAL LOADLEVELER SETTINGS
[622]	614	execute_in_shell="#!/bin/ksh"
	615	use_shell="# @ shell = /bin/ksh"
	616	consumable_memory="ConsumableMemory($memory mb)"
	617	class="# @ class = $queue"
	618	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	619	network_to_use="# @ network.mpi = sn_all,shared,us"
	620	data_limit="# @ data_limit = 1.76gb"
	621	image_size="# @ image_size = 50"
[693]	622	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	623
[693]	624	if [[ $email_notification = none ]]
	625	then
	626	notify_user=""
	627	else
	628	notify_user="# @ notify_user = $email_notification"
	629	if [[ $delete_dayfile = true ]]
	630	then
	631	notification='# @ notification = never'
	632	fi
	633	fi
[622]	634
[312]	635	if [[ $remote_host = ibmh ]]
[1]	636	then
[312]	637	data_limit=""
	638	network_to_use=""
[1184]	639	class="# @ class = $queue"
[312]	640	environment=""
[814]	641	rset="# @ rset = RSET_MCM_AFFINITY"
	642	task_affinity="# @ task_affinity = core(1)"
[693]	643	elif [[ $remote_host = ibmkisti ]]
	644	then
	645	network_to_use="# @ network.MPI = sn_all,shared,US"
	646	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	647	if [[ $threads_per_task = 1 ]]
	648	then
	649	rset="# @ rset = RSET_MCM_AFFINITY"
	650	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	651	fi
[693]	652	environment=""
	653	use_shell=""
	654	data_limit=""
	655	image_size=""
[622]	656	elif [[ $remote_host = ibmku ]]
	657	then
	658	execute_in_shell="#!/usr/bin/ksh"
	659	use_shell="# @ shell = /usr/bin/ksh"
	660	consumable_memory=""
	661	environment=""
	662	network_to_use="# @ network.mpi = sn_all,shared,us"
	663	data_limit=""
	664	image_size=""
	665	elif [[ $remote_host = ibms ]]
	666	then
	667	network_to_use="# @ network.mpi = csss,shared,us"
[1]	668	fi
	669
	670	cat > $job_to_send << %%END%%
[622]	671	$execute_in_shell
	672	$use_shell
[1]	673
	674	# @ job_type = parallel
[693]	675	# @ job_name = $job_name
[1]	676	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	677	# @ output = $remote_dayfile
	678	# @ error = $remote_dayfile
[693]	679	$wall_clock_limit
[622]	680	$image_size
[312]	681	$class
	682	$environment
[1]	683	$network_to_use
	684	$data_limit
[693]	685	$rset
	686	$mcm_affinity_options
[814]	687	$task_affinity
[1]	688	$notification
[693]	689	$notify_user
[1]	690
	691	%%END%%
	692
	693	if (( nodes > 0 ))
	694	then
	695
[693]	696	if [[ $remote_host != ibmkisti ]]
	697	then
	698
	699	cat >> $job_to_send << %%END%%
[1]	700	# @ node = $nodes
[622]	701	# @ tasks_per_node = $processes_per_node
[1]	702	# @ node_usage = $node_usage
	703	# @ queue
	704
	705	%%END%%
	706
[693]	707	else
	708
	709	cat >> $job_to_send << %%END%%
[696]	710	# @ total_tasks = $mpi_tasks
[693]	711	# @ blocking = unlimited
	712	# @ queue
	713
	714	%%END%%
	715
	716	fi
	717
[1]	718	else
	719
[1289]	720	cat >> $job_to_send << %%END%%
[1]	721	# @ blocking = unlimited
	722	# @ total_tasks = $numprocs
	723	# @ node_usage = $node_usage
	724	# @ queue
	725
	726	%%END%%
	727
	728	fi
	729
[1289]	730	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
[696]	731	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	732	then
	733	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	734	fi
	735
[1]	736	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	737	then
	738
	739	cat > $job_to_send << %%END%%
	740	#!/bin/ksh
	741
	742	# @ job_type = serial
	743	# @ node_usage = $node_usage
	744	# @ job_name = palm
	745	# @ wall_clock_limit = ${timestring},$timestring
	746	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	747	# @ output = $remote_dayfile
	748	# @ error = $remote_dayfile
[312]	749	$class
[1]	750	$notification
	751
	752	# @ queue
	753
	754	%%END%%
	755
[1274]	756	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1224]	757	then
	758
	759	if [[ $numprocs != 0 ]]
	760	then
	761	cat > $job_to_send << %%END%%
[1255]	762	#!/bin/bash
[1224]	763	#PBS -N $job_name
	764	#PBS -l walltime=$timestring
[1264]	765	#PBS -l nodes=$nodes:ppn=$processes_per_node
[1224]	766	#PBS -o $remote_dayfile
	767	#PBS -j oe
	768	#PBS -q $queue
	769
	770	$module_calls
	771
	772	%%END%%
	773
	774	else
	775
	776	continue
	777
	778	fi
	779
[1197]	780	elif [[ $remote_host = lccrayf ]]
	781	then
	782
	783	if [[ $numprocs != 0 ]]
	784	then
	785	cat > $job_to_send << %%END%%
	786	#!/bin/bash -l
	787	#SBATCH -J $job_name
	788	#SBATCH -t $timestring
	789	#SBATCH -N $nodes
	790	#SBATCH --ntasks-per-node=$processes_per_node
	791	#SBATCH -p $queue
	792	#SBATCH -o $remote_dayfile
	793	#SBATCH -e $remote_dayfile
	794
	795	$init_cmds
	796	$module_calls
	797
	798	%%END%%
	799
	800	else
	801	cat > $job_to_send << %%END%%
	802	#!/bin/bash -l
	803	#SBATCH -J $job_name
	804	#SBATCH -t $timestring
	805	#SBATCH -l ncpus=1
	806	#SBATCH -l pmem=${memory}mb
	807	#SBATCH -m abe
	808	#SBATCH -o $remote_dayfile
	809	#SBATCH -e $remote_dayfile
	810
	811	$init_cmds
	812	$module_calls
	813
	814	%%END%%
	815
	816	fi
	817
[1099]	818	elif [[ $remote_host = lcflow ]]
	819	then
[1202]	820	if [ $memory -gt 1800 ]; then
[1099]	821	use_himem=""
	822	else
	823	use_himem="#"
	824	fi
	825	if [[ $numprocs != 0 ]]
	826	then
[1202]	827	pe_set="#$ -pe impi41 $numprocs"
[1099]	828	else
[1202]	829	pe_set="#$ -pe impi41 1"
[1099]	830	fi
	831	if [[ $queue = default ]]
	832	then
	833	queue_set=""
	834	else
	835	queue_set="#$ -q $queue"
	836	fi
	837	[[ "$disc_space" = "" ]] && disc_space=50
	838
	839	cat > $job_to_send << %%END%%
	840	#!/bin/bash
	841	#$ -S /bin/bash
	842	#$ -N $job_name
	843	#$ -cwd
	844	#$ -l h_rt=$timestring
	845	#$ -l h_vmem=${memory}M
	846	#$ -o $remote_dayfile
	847	#$ -j y
	848	$pe_set
	849	#$ -R y
	850	${use_himem}#$ -l highmem=true
	851	#$ -l h_fsize=${disc_space}G
	852	$queue_set
	853
	854	%%END%%
	855
[1021]	856	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
[368]	857	then
	858
	859	if [[ $numprocs != 0 ]]
	860	then
	861	cat > $job_to_send << %%END%%
	862	#!/bin/ksh
	863	#PBS -N $job_name
	864	#PBS -l walltime=$timestring
	865	#PBS -l ncpus=$numprocs
	866	#PBS -l pmem=${memory}mb
	867	#PBS -o $remote_dayfile
[1021]	868	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[368]	869	#PBS -j oe
	870
	871	mpd &
	872
	873	%%END%%
	874
	875	else
	876	cat > $job_to_send << %%END%%
	877	#!/bin/ksh
	878	#PBS -N $job_name
	879	#PBS -l walltime=$timestring
	880	#PBS -l ncpus=1
	881	#PBS -l pmem=${memory}mb
	882	#PBS -o $remote_dayfile
	883	#PBS -j oe
	884
	885	%%END%%
	886
	887	fi
	888
[1040]	889	elif [[ $remote_host = lckiaps ]]
	890	then
	891
	892	if [[ $numprocs != 0 ]]
	893	then
	894	cat > $job_to_send << %%END%%
	895	#!/bin/ksh
	896	#PBS -N $job_name
	897	#PBS -l walltime=$timestring
	898	#PBS -l select=1:ncpus=$numprocs
	899	#PBS -l pmem=${memory}mb
	900	#PBS -q $queue
	901	#PBS -o $remote_dayfile
	902	#PBS -j oe
	903	#PBS -V
	904
	905	%%END%%
	906
	907	else
	908	cat > $job_to_send << %%END%%
	909	#!/bin/ksh
	910	#PBS -N $job_name
	911	#PBS -l walltime=$timestring
	912	#PBS -l ncpus=1
	913	#PBS -l pmem=${memory}mb
	914	#PBS -o $remote_dayfile
	915	#PBS -j oe
	916
	917	%%END%%
	918
	919	fi
	920
[693]	921	elif [[ $remote_host = lcyon ]]
	922	then
	923
	924	if [[ $numprocs != 0 ]]
	925	then
	926	cat > $job_to_send << %%END%%
	927	#!/bin/ksh
	928	#PBS -N $job_name
	929	#PBS -l walltime=$timestring
	930	#PBS -l ncpus=$numprocs
	931	#PBS -l pmem=${memory}mb
	932	#PBS -o $remote_dayfile
	933	#PBS -j oe
	934
	935	%%END%%
	936
	937	else
	938	cat > $job_to_send << %%END%%
	939	#!/bin/ksh
	940	#PBS -N $job_name
	941	#PBS -l walltime=$timestring
	942	#PBS -l ncpus=1
	943	#PBS -l pmem=${memory}mb
	944	#PBS -o $remote_dayfile
	945	#PBS -j oe
	946
	947	%%END%%
	948
	949	fi
	950
[892]	951	elif [[ $remote_host = lcxe6 ]]
[164]	952	then
	953
	954	if [[ $numprocs != 0 ]]
	955	then
	956	cat > $job_to_send << %%END%%
[799]	957	#!/bin/ksh
[164]	958	#PBS -S /bin/ksh
	959	#PBS -N $job_name
[552]	960	#PBS -A $project_account
[206]	961	#PBS -j oe
[164]	962	#PBS -l walltime=$timestring
	963	#PBS -l mppwidth=${numprocs}
[622]	964	#PBS -l mppnppn=${processes_per_node}
[164]	965	#PBS -m abe
	966	#PBS -o $remote_dayfile
[492]	967	$email_directive
[164]	968
[892]	969	$init_cmds
[493]	970	$module_calls
[343]	971
[164]	972	%%END%%
	973
[108]	974	else
	975	cat > $job_to_send << %%END%%
	976	#!/bin/ksh
[168]	977	#PBS -S /bin/ksh
[108]	978	#PBS -N $job_name
[552]	979	#PBS -A $project_account
[206]	980	#PBS -j oe
[108]	981	#PBS -l walltime=$timestring
	982	#PBS -l ncpus=1
	983	#PBS -l pmem=${memory}mb
	984	#PBS -m abe
[492]	985	$email_directive
[108]	986	#PBS -o $remote_dayfile
	987
[892]	988	$init_cmds
[493]	989	$module_calls
[343]	990
[108]	991	%%END%%
	992
	993	fi
	994
[440]	995	elif [[ $remote_host = lckyoto ]]
[437]	996	then
	997
[440]	998	cat > $job_to_send << %%END%%
[799]	999	#!/bin/ksh
[440]	1000	# @\$-o $remote_dayfile
	1001	# @\$-eo -oi
	1002	# @\$-lP 16
[799]	1003	# @\$-lp 1
[440]	1004	# @\$-lm 28gb -llm unlimited -ls unlimited
	1005	# @\$-q $queue
	1006	# @\$-Pvn abs_pack
	1007	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	1008	#. /thin/local/etc/setprofile/intel-11.0.sh
	1009	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1010	. ~/.myprofile
	1011	#. /home2/t/t51254/palm/current_version/myprofile
	1012	#. /thin/apps/pgi/mpi.sh
	1013	#
	1014	env
	1015	#
	1016	set -x
	1017
	1018	%%END%%
	1019
	1020	elif [[ $remote_host = lcxt5m ]]
	1021	then
	1022
[437]	1023	if [[ $numprocs != 0 ]]
	1024	then
	1025	cat > $job_to_send << %%END%%
[799]	1026	#!/bin/ksh
[437]	1027	#PBS -S /bin/ksh
	1028	#PBS -N $job_name
	1029	#PBS -j oe
	1030	#PBS -l walltime=$timestring
	1031	#PBS -l mppwidth=${numprocs}
[622]	1032	#PBS -l mppnppn=${processes_per_node}
[437]	1033	#PBS -m abe
	1034	#PBS -o $remote_dayfile
	1035
[892]	1036	$init_cmds
[493]	1037	$module_calls
[437]	1038
	1039	%%END%%
	1040
	1041	else
	1042	cat > $job_to_send << %%END%%
	1043	#!/bin/ksh
	1044	#PBS -S /bin/ksh
	1045	#PBS -N $job_name
	1046	#PBS -j oe
	1047	#PBS -l walltime=$timestring
	1048	#PBS -l ncpus=1
	1049	#PBS -l pmem=${memory}mb
	1050	#PBS -m abe
	1051	#PBS -o $remote_dayfile
	1052
[892]	1053	$init_cmds
[493]	1054	$module_calls
[437]	1055
	1056	%%END%%
	1057
	1058	fi
	1059
[1099]	1060	elif [[ $remote_host = lckyuh ]]
	1061	then
	1062	cat > $job_to_send << %%END%%
	1063	#!/bin/bash
	1064	#PJM -L "rscgrp=$queue"
	1065	#PJM -L "node=$nodes"
	1066	#PJM --mpi "proc=$numprocs"
	1067	#PJM -L "elapse=$timestring"
	1068	#PJM -o $remote_dayfile
	1069	#PJM -j
	1070	#PJM -X
	1071	#PJM --no-stging
	1072
	1073	export LANG=en_US.UTF-8
	1074	%%END%%
	1075
[1090]	1076	elif [[ $remote_host = lckyut ]]
	1077	then
	1078	cat > $job_to_send << %%END%%
	1079	#!/bin/bash
[1099]	1080	#PJM -L "rscgrp=$queue"
[1090]	1081	#PJM -L "vnode=$numprocs"
	1082	#PJM -L "vnode-core=1"
	1083	#PJM -L "elapse=$timestring"
	1084	#PJM --mpi proc=$numprocs
	1085	#PJM -o $remote_dayfile
	1086	#PJM -j
[1099]	1087	#PJM -X
	1088	#PJM --no-stging
[1090]	1089
	1090	export LANG=en_US.UTF-8
	1091	%%END%%
	1092
[1]	1093	elif [[ $remote_host = nech ]]
	1094	then
	1095
	1096	if (( nodes > 1 ))
	1097	then
	1098	cat > $job_to_send << %%END%%
	1099	#!/bin/ksh
[622]	1100	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1101	#PBS -l ${qsubmem}=${Memory}gb
	1102	#PBS -b $nodes
	1103	#PBS -o $remote_dayfile
	1104	#PBS -N palm
	1105	#PBS -j o
	1106	#PBS -T mpisx
	1107
	1108	%%END%%
	1109
	1110	elif [[ $numprocs != 0 ]]
	1111	then
	1112	cat > $job_to_send << %%END%%
	1113	#!/bin/ksh
[622]	1114	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1115	#PBS -l ${qsubmem}=${Memory}gb
	1116	#PBS -o $remote_dayfile
	1117	#PBS -N palm
	1118	#PBS -j o
	1119
	1120	%%END%%
	1121
	1122	else
	1123	cat > $job_to_send << %%END%%
	1124	#!/bin/ksh
	1125	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1126	#PBS -o $remote_dayfile
	1127	#PBS -j o
	1128
	1129	%%END%%
	1130
	1131	fi
	1132
	1133	elif [[ $remote_host = lctit ]]
	1134	then
	1135	cat > $job_to_send << %%END%%
[635]	1136	#!/bin/ksh
[892]	1137	$init_cmds
[678]	1138	$module_calls
	1139
[1]	1140	%%END%%
	1141
[1289]	1142	# SET OPTIONS FOR SUBMIT-COMMAND
[678]	1143	if [[ $tasks_per_node != $processes_per_node ]]
	1144	then
[1378]	1145	submcom="$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 $queue "
[678]	1146	else
[1378]	1147	submcom="$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]	1148	fi
[1]	1149
	1150	else
	1151
	1152	cat > $job_to_send << %%END%%
	1153	# @\$-q ${queue}
	1154	# @\$-l${qsubtime} $timestring
	1155	# @\$-l${qsubmem} ${memory}mb
	1156	# @\$-o $remote_dayfile
	1157	# @\$-eo
	1158
	1159	%%END%%
	1160
	1161	fi
	1162
	1163
[1289]	1164	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
	1165	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
	1166	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
[1]	1167	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1168	then
	1169	echo "set +vx" >> $job_to_send
	1170	echo "trap '" >> $job_to_send
	1171	echo "set +vx" >> $job_to_send
[1289]	1172	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $remote_host = lccrayb \|\| $remote_host = lccrayh \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lcflow \|\| $remote_host = lckiaps \|\| $remote_host = lckyu* ]]
[1]	1173	then
[622]	1174	if [[ $remote_host = ibmh ]]
[1]	1175	then
	1176	return_queue=c1
[693]	1177	elif [[ $remote_host = ibmkisti ]]
	1178	then
	1179	return_queue=class.1-2
[622]	1180	elif [[ $remote_host = ibmku ]]
	1181	then
	1182	return_queue=sdbg2
[1]	1183	elif [[ $remote_host = ibms ]]
	1184	then
	1185	return_queue=p_normal
[1274]	1186	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1187	then
	1188	return_queue=dataq
[1040]	1189	elif [[ $remote_host = lckiaps ]]
	1190	then
	1191	return_queue=express
[1099]	1192	elif [[ $remote_host = lckyuh ]]
	1193	then
	1194	return_queue=cx-single
[1090]	1195	elif [[ $remote_host = lckyut ]]
	1196	then
	1197	return_queue=cx-single
[1]	1198	else
	1199	return_queue=unknown
	1200	fi
	1201
	1202	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1203	then
	1204
[622]	1205	if [[ $remote_host = ibmku ]]
	1206	then
[1289]	1207	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
	1208	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1209	else
[1289]	1210	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
[622]	1211	fi
[1289]	1212	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
	1213	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
	1214	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
	1215	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
	1216	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
	1217	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
[312]	1218	if [[ $host != "ibmh" ]]
	1219	then
[1289]	1220	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
[312]	1221	fi
[1289]	1222	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
	1223	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
[1]	1224
[1289]	1225	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
	1226	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
[1]	1227
[1289]	1228	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
	1229	echo "echo \"batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$identifier" >> $job_to_send
[622]	1230	if [[ $remote_host = ibmku ]]
	1231	then
[1289]	1232	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
[622]	1233	fi
[1289]	1234	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
[1]	1235
	1236	elif [[ $remote_host = nech ]]
	1237	then
	1238	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1289]	1239	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1240	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1241	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1242	echo "#PBS -j o" >> $job_to_send
	1243	echo " " >> $job_to_send
	1244	echo "set -x" >> $job_to_send
	1245	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
[1094]	1246	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1]	1247	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1248	echo "%%END%%" >> $job_to_send
	1249
[1099]	1250	elif [[ $remote_host = lckyuh ]]
	1251	then
[1289]	1252	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1099]	1253	echo "#!/bin/bash" >> $job_to_send
	1254	echo "#PJM -L \"node=1\"" >> $job_to_send
	1255	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1256	echo "#PJM --no-stging" >> $job_to_send
	1257	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1258	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1259	echo "#PJM -j" >> $job_to_send
	1260	echo " " >> $job_to_send
	1261	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1262	echo "set -x" >> $job_to_send
	1263	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1264	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1265	echo "%%END%%" >> $job_to_send
	1266
[1090]	1267	elif [[ $remote_host = lckyut ]]
	1268	then
[1289]	1269	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1090]	1270	echo "#!/bin/bash" >> $job_to_send
	1271	echo "#PJM -L \"vnode=1\"" >> $job_to_send
	1272	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
	1273	echo "#PJM --no-stging" >> $job_to_send
	1274	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
	1275	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1276	echo "#PJM -j" >> $job_to_send
	1277	echo " " >> $job_to_send
	1278	echo "export LANG=en_US.UTF-8" >> $job_to_send
	1279	echo "set -x" >> $job_to_send
[1094]	1280	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1090]	1281	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1282	echo "%%END%%" >> $job_to_send
	1283
[1274]	1284	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1285	then
[1289]	1286	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1255]	1287	echo "#!/bin/bash" >> $job_to_send
	1288	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	1289	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[1262]	1290	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[1255]	1291	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1292	echo "#PBS -j oe" >> $job_to_send
	1293	echo " " >> $job_to_send
	1294	echo "set -x" >> $job_to_send
	1295	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1296	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1297	echo "%%END%%" >> $job_to_send
	1298
[1099]	1299	elif [[ $remote_host = lcflow ]]
	1300	then
[1289]	1301	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
[1099]	1302	echo "#!/bin/bash" >> $job_to_send
	1303	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
	1304	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
	1305	echo "SGEPREFIX -cwd" >> $job_to_send
	1306	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
[1202]	1307	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
	1308	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
[1099]	1309	echo "SGEPREFIX -j y" >> $job_to_send
[1289]	1310	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
[1099]	1311	echo " " >> $job_to_send
	1312	echo "set -x" >> $job_to_send
	1313	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
	1314	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
	1315	echo "" >> $job_to_send
	1316	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1317	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
[1289]	1318	echo "rm -f scpjob.${identifier}" >> $job_to_send
[1099]	1319	echo "%%END%%" >> $job_to_send
[1289]	1320	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
	1321	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
[1099]	1322
[1]	1323	else
	1324
[1289]	1325	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
[1]	1326	echo "# @\\\$-q $return_queue" >> $job_to_send
	1327	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1328	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1329	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1330	then
	1331	echo "# @\$-l mpp_p=0" >> $job_to_send
	1332	fi
	1333	echo '# @\$-lF 10mb' >> $job_to_send
	1334	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1335	echo '# @\\\$-eo' >> $job_to_send
	1336	echo " " >> $job_to_send
	1337	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1338	then
	1339	echo "set +vx" >> $job_to_send
	1340	echo ". .profile" >> $job_to_send
	1341	fi
	1342	echo "set -x" >> $job_to_send
[1094]	1343	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
[1]	1344	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1345	echo "%%END%%" >> $job_to_send
	1346	fi
	1347
	1348	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1349	then
[1289]	1350	echo "llsubmit scpjob.$identifier" >> $job_to_send
[1274]	1351	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
[1255]	1352	then
[1289]	1353	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
[1]	1354	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1355	then
[1289]	1356	echo "qsub -J n scpjob.$identifier" >> $job_to_send
[1]	1357	elif [[ $remote_host = t3es ]]
	1358	then
[1289]	1359	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
[1043]	1360	elif [[ $remote_host = lckiaps ]]
	1361	then
[1289]	1362	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
	1363	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
	1364	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
[1099]	1365	elif [[ $remote_host = lckyu* ]]
[1090]	1366	then
[1289]	1367	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
	1368	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
[1099]	1369	elif [[ $remote_host = lcflow ]]
	1370	then
[1289]	1371	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
	1372	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
[1]	1373	else
[1289]	1374	echo "$submcom scpjob.$identifier" >> $job_to_send
[1]	1375	fi
[1043]	1376	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
[622]	1377	then
[1289]	1378	echo "rm scpjob.$identifier" >> $job_to_send
[622]	1379	fi
[1]	1380	if [[ $remote_host = nech ]]
	1381	then
	1382	echo "cd -" >> $job_to_send
	1383	fi
	1384	else
	1385	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
[1289]	1386	# ??? funktioniert das ÃŒberhaupt noch ???
[1]	1387	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1388	fi
	1389	echo "set -x" >> $job_to_send
	1390	echo " ' exit" >> $job_to_send
	1391	echo "set -x" >> $job_to_send
	1392	fi
	1393
	1394
	1395
[1289]	1396	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
[1]	1397	cat $file_to_send >> $job_to_send
[69]	1398
[1]	1399	if [[ $remote_host = ibm ]]
	1400	then
	1401	echo " " >> $job_to_send
	1402	echo "exit" >> $job_to_send
	1403	fi
[635]	1404
[1289]	1405	# REMOVE JOB-FILE
[1099]	1406	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
[69]	1407	then
	1408	echo " " >> $job_to_send
	1409	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1410	fi
[1]	1411
	1412
[1289]	1413	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
[1]	1414	if [[ $no_submit = false ]]
	1415	then
	1416	if [[ $remote_host != $local_host ]]
	1417	then
	1418	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[1289]	1419	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
[1]	1420	then
[82]	1421	job_catalog_save=$job_catalog
	1422	job_catalog=job_queue
	1423	elif [[ $remote_host = nech ]]
	1424	then
	1425	job_catalog_save=$job_catalog
	1426	job_catalog=/hpf/b/${remote_user}/job_queue
	1427	fi
[1096]	1428	if [[ $remote_host = nech ]]
[82]	1429	then
[1289]	1430	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
[1094]	1431	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1432	else
[1094]	1433	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1434	fi
[82]	1435	if [[ $? = 1 ]]
	1436	then
	1437	locat=scp; exit
	1438	fi
	1439	if [[ $remote_host = ibms ]]
	1440	then
	1441	job_catalog=$job_catalog_save
	1442	fi
[1]	1443	[[ $verify = true ]] && printf "\n >>> finished\n"
	1444	else
	1445	eval job_catalog=$job_catalog
	1446	cp $job_to_send ${job_catalog}/$job_on_remhost
	1447	fi
	1448
	1449
	1450
[1289]	1451	# START NQS- / LOADLEVELER-JOB
[1]	1452	if [[ $remote_host != $local_host ]]
	1453	then
	1454	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1455
[1289]	1456	if [[ $remote_host = ibmku ]]
[1]	1457	then
[1094]	1458	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1099]	1459	elif [[ $remote_host = lcflow ]]
	1460	then
[1202]	1461	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
[1]	1462	else
[1094]	1463	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1464	fi
[1]	1465
	1466	[[ $verify = true ]] && printf " >>> o.k.\n"
	1467	else
	1468	cd $job_catalog
[1289]	1469	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
[1]	1470	then
	1471	eval $submcom $job_on_remhost
[1289]	1472	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
[108]	1473	then
[635]	1474	chmod u+x $job_on_remhost
[108]	1475	eval $submcom $job_on_remhost
[1]	1476	elif [[ $local_host = nech ]]
	1477	then
	1478	if [[ $queue = default ]]
	1479	then
[799]	1480	eval $submcom $job_on_remhost
[1]	1481	else
[799]	1482	eval $submcom -q $queue $job_on_remhost
[1]	1483	fi
	1484	else
	1485	qsub $job_on_remhost
	1486	fi
[622]	1487
[1289]	1488	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
	1489	# AT THE END OF THE JOB
[1099]	1490	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
[622]	1491	then
	1492	rm $job_on_remhost
	1493	fi
[1]	1494	cd - > /dev/null
	1495	fi
	1496	fi
	1497
	1498
	1499
[1289]	1500	# FINAL ACTIONS
[1]	1501	if [[ $no_submit = false ]]
	1502	then
[1099]	1503	rm -f $job_to_send
[1]	1504	fi
[1266]	1505	[[ $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 |