Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1046

Last change on this file since 1046 was 1046, checked in by maronga, 12 years ago
put scripts and utilities under GPL
Property svn:keywords set to `Id Rev`
File size: 52.9 KB

Rev	Line
[1]	1	#!/bin/ksh
[1046]	2	#--------------------------------------------------------------------------------#
	3	# This file is part of PALM.
	4	#
	5	# PALM is free software: you can redistribute it and/or modify it under the terms
	6	# of the GNU General Public License as published by the Free Software Foundation,
	7	# either version 3 of the License, or (at your option) any later version.
	8	#
	9	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
	10	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
	11	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
	12	#
	13	# You should have received a copy of the GNU General Public License along with
	14	# PALM. If not, see <http://www.gnu.org/licenses/>.
	15	#
	16	# Copyright 1997-2012 Leibniz University Hannover
	17	#--------------------------------------------------------------------------------#
	18	#
	19	# Current revisions:
	20	# -----------------
	21	# code put under GPL (PALM 3.9)
	22	#
	23	# Former revisions:
	24	# -----------------
[169]	25	# $Id: subjob 1046 2012-11-09 14:38:45Z maronga $
[1]	26
[1046]	27	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
[1]	28	# Prozedur zum automatischen Generieren von Batch-Jobs, die unter NQS
	29	# laufen sollen und deren Ergebnis (Dayfile) zum Job-generierenden
	30	# Host zurueckgeschickt werden sollen
	31
	32
	33	# letzte Aenderung:
	34	# 29/06/94 - Siggi - Beginn mit Erstellung der Originalversion
	35	# 08/07/94 - Siggi - Originalversion abgeschlossen (Version 1.0)
	36	# 06/02/98 - Siggi - berte validiert
	37	# 27/01/01 - Siggi - ground.yonsei.ac.kr validiert, Jobs zur T3E in Korea
	38	# moeglich
	39	# 08/02/01 - Siggi - alle subjob-Meldungen ins englische uebersetzt
	40	# 25/05/02 - Siggi - Unterstuetzung des LoadLeveler
	41	# 30/05/02 - Siggi - Validierung fuer ibm-Rechner in Seoul (nobel) sowie
	42	# allgemeine Anpassungen fuer ibm-Rechner
	43	# 15/10/02 - Siggi - Neue Default-Jobklasse (p_normal) fuer IBM in Seoul
	44	# Ruecktransfer des Jobprotokolls fuer diese
	45	# Maschine abgeschaltet
	46	# 31/10/02 - Siggi - berni validiert
	47	# 06/11/02 - Siggi - Neue Jobklassen auf ibmb und ibmh
	48	# 08/11/02 - Siggi - quanero validiert
	49	# 11/12/02 - Siggi - Notification fuer Transfer-Jobs abgeschaltet
	50	# 23/01/03 - Siggi - hostname nobel changed to nobela
	51	# 06/02/03 - Siggi - gregale validated
	52	# 12/02/03 - Siggi - orkan and maestro validated
	53	# 21/02/03 - Siggi - all nobel nodes in Seoul validated
	54	# 12/03/03 - Siggi - nec at DKRZ validated
	55	# 13/03/03 - Siggi - new nqs resource variable Memory
	56	# 07/04/03 - Siggi - processor request option -c on nech needs tasks per
	57	# node
	58	# 11/04/03 - Siggi - network on ibms has attribute "shared"
	59	# 31/07/03 - Siggi - nqs2 on nech implemented (provisional: -h nech2)
	60	# cxxl added to ibmh
	61	# 29/08/03 - Siggi - changes in job queues and communication system on
	62	# ibms
	63	# 24/10/03 - Siggi - using alternate hanni address 130.75.4.2
	64	# 30/10/03 - Siggi - nech is not supported any more
	65	# 10/11/03 - Siggi - nech2 renamed to nech
	66	# 20/11/03 - Siggi - submit command on nech changed from qsub.test to qsub
	67	# 29/03/04 - Siggi - ground not supported any more, gfdl3 validated
	68	# 31/03/04 - Siggi - new option -N for node usage
	69	# 12/04/04 - Siggi - scp2 instead of scp used for transfer from decalpha
	70	# due to error in ssh installation (otherwise a prompt
	71	# for the password appears)
	72	# 23/07/04 - Siggi - changes due to the new berni configuration
	73	# (federation switch)
	74	# 01/09/04 - Gerald new job-classes on hanni
	75	# 08/09/04 - Siggi - hanni IP address changed to 130.75.4.10
	76	# 23/11/04 - Siggi - new job class cdata on hanni and berni
	77	# 03/12/04 - Siggi - notification on ibm switched of in case of
	78	# delete_dayfile = true, node usage in cdev set to
	79	# shared
	80	# 16/02/05 - Gerald hababai validated
	81	# 29/03/05 - Micha - new job class channi on hanni
	82	# 11/05/05 - Siggi - ConsumableMemory is now required as resource keyword
	83	# on ibms
	84	# 24/05/05 - Siggi - Default queue on ibms changed from p_normal_1.3 to
	85	# p_normal
	86	# 30/06/05 - Siggi - network changed for queue cdev from "us" to "ip"
	87	# 12/07/05 - Siggi - in network.mpi on ibmh/ibmb "csss" changed to
	88	# "sn_all", new job class cexp
	89	# 08/09/05 - Siggi - IP-address of gfdl3 changed
	90	# 31/10/05 - Siggi - new job class pp on hurricane, serial jobs on
	91	# hurricane (with -X0)
	92	# 01/11/05 - Siggi - missing queue for jobs submitted on nech (for nech)
	93	# added
	94	# 30/12/05 - Siggi - change of IP adresses in subnet 130.75.105
	95	# 09/02/06 - Siggi - ibmy admitted
	96	# 10/02/06 - Siggi - scp2 changed to /bin/scp on decalpha
	97	# 13/04/06 - Siggi - ostria admitted
	98	# 18/04/06 - Siggi - new option -O for OpenMP usage
	99	# 24/05/06 - Siggi - lctit admitted, ftpjob renamed scpjob
	100	# 25/07/06 - Siggi - gfdl5 (ibmy) admitted for submitting jobs
	101	# 27/09/06 - Siggi - breg/hreg extended with berni/hanni
	102	# 25/10/06 - Siggi - data_limit set to 1.76 GByte on hanni and berni
	103	# 28/11/06 - Siggi - levanto admitted
[5]	104	# 13/02/07 - Siggi - hpmuk releated code removed
[35]	105	# 01/03/07 - Siggi - adjustments for RIAM machines gate and NEC-SX8 (n-sx)
[82]	106	# 12/04/07 - Siggi - option -f (filetransfer protocol) removed, scp only
[102]	107	# 27/07/07 - Siggi - autan admitted
[108]	108	# 03/08/07 - Marcus- lcfimm admitted
[113]	109	# 08/10/07 - Siggi - further job classes added for hanni (csoloh...)
[118]	110	# 15/10/07 - Siggi - preliminary adjustments for lctit based on Jin's
	111	# suggestions
[122]	112	# 19/10/07 - Marcus- add new optional argument -g group_number
[125]	113	# 19/10/07 - Siggi - a ";" was missing in the last change done by Marcus
[129]	114	# 30/10/07 - Marcus- further adjustments for queues on lctit
[164]	115	# 15/05/08 - Siggi - adjustments for lcxt4 (Bergen Center for Computational
	116	# Science)
[179]	117	# 14/07/08 - Siggi - adjustments for lcsgih
[205]	118	# 23/09/08 - Gerald- paesano admitted
[206]	119	# 02/10/08 - Siggi - PBS adjustments for lcxt4
[251]	120	# 02/03/09 - Siggi - Adjustments for new NEC-SX9 at RIAM
[292]	121	# 16/04/09 - Marcus- Adjustments for lcsgib and lcsgih
[312]	122	# 21/04/09 - Siggi - adjustments for new IBM at DKRZ, which is now ibmh
[325]	123	# 18/05/09 - Siggi - Settings for serial jobs on lcsgi changed
[343]	124	# 24/06/09 - BjornM- adjustments for lcxt4 (loading modules manually)
[352]	125	# 08/07/09 - Siggi - option -e added (email notification on lcsgih/b)
[366]	126	# 20/07/09 - Siggi - On lcsgi, jobs for returning the job protocol are
	127	# now run on the data nodes (feature=data)
[367]	128	# 25/08/09 - BjornM- adapted for lck
[372]	129	# 26/08/09 - Marcus- caurus admitted; optional qos feature hiprio on lcsgi
[376]	130	# 03/09/09 - Siggi - PBS sgi feature directive only used if explicitly
	131	# set in the config file by the user
[416]	132	# 16/10/09 - Carolin-adjustments for archiving on SGI-ICE of binary files;
	133	# adding special1q
[407]	134	# 01/12/09 - BjornM- re-adjustments for lcxt4, added sno (130.75.105.113)
[416]	135	# 16/10/09 - Carolin-adding permq
[440]	136	# 01/02/10 - Siggi - adapted for lcxt5m and lckyoto (Fujitsu HX600)
[475]	137	# 03/02/10 - Siggi - bug in serial jobs removed
[492]	138	# 26/02/10 - BjornM- re-adjustments for lcxt4 (new modules, email
	139	# notification)
[493]	140	# 01/03/10 - Siggi - loading of modules controlled by environment variable
	141	# module_calls
[551]	142	# 17/08/10 - BjornM- re-adjustments for lcxt4 (location of qsub)
	143	# 25/08/10 - BjornM- account geofysisk replaced by guest for lcxt4
[552]	144	# 25/08/10 - Siggi - new variable project_account in pbs-statements for
	145	# lcxt4
[622]	146	# 08/12/10 - Siggi - initialization of the module command changed for
	147	# SGI-ICE/lcsgi
	148	# adjustments for Kyushu Univ. (lcrte, ibmku)
[635]	149	# 14/12/10 - Siggi - adjustments for new Tsubame system at Tokyo
	150	# institute of technology (lctit)
[678]	151	# 02/02/11 - Siggi - further asjustments for Tsubame concerning openMP
[693]	152	# 06/03/11 - Siggi - adjustments for ibmkisti
[696]	153	# 17/03/11 - Siggi - adjustments for openmp usage on ibmkisti
[717]	154	# 03/04/11 - Micha - added lckordi
[739]	155	# 17/08/11 - Siggi - hicegate0 added
[748]	156	# 18/08/11 - Siggi - workaround on lcsgi in order to avoid appends to file
	157	# last_job_transfer_protocol
[749]	158	# 21/08/11 - Siggi - inferno admitted
[789]	159	# 29/11/11 - Siggi - adjustments for lcsgih/lcsgib queues, pingui admitted
[800]	160	# 21/12/11 - Theres- solano admitted
[892]	161	# 31/01/12 - Matthias - adjustments for ibmh
	162	# 02/05/12 - Bjoern - lcxt4 is now lcxe6, adjustments for lcxe6,
	163	# bugfixes for the parameter file check, initial
	164	# commands on ssh calls sourced out to .mrun.config
[1021]	165	# 02/10/12 - Siggi - adjustments for lcsb
[1040]	166	# 01/11/12 - Siggi - adjustments for lckiaps
[1]	167
[437]	168
[1021]	169
[1]	170	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
[352]	171	delete_dayfile=false
[799]	172	email_notification=none
[122]	173	group_number=none
[1]	174	locat=normal
	175	no_default_queue=none
	176	no_submit=false
	177	job_catalog="~/job_queue"
	178	job_name=none
	179	local_user=$LOGNAME
	180	node_usage=shared
[475]	181	numprocs=0
[1]	182	punkte="..........................................................."
	183	submcom=qsub
	184	queue=default
	185	remote_host=none
	186	remote_user=""
	187	verify=true
	188
	189	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	190	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	191	typeset -L20 spalte1
	192	typeset -R40 spalte2
	193	typeset -L60 spalte3
	194
	195
	196
	197	# FEHLERBEHANDLUNG
	198	# BEI EXIT:
	199	trap 'if [[ $locat != normal ]]
	200	then
	201	case $locat in
	202	(option) printf "\n --> available optios can be displayed"
	203	printf " by typing:"
	204	printf "\n \"subjob ?\" \n";;
	205	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	206	(*) printf "\n +++ unknown error"
	207	printf "\n please inform S. Raasch!\n"
	208	esac
	209	[[ -f $job_to_send ]] && rm $job_to_send
	210	printf "\n\n+++ SUBJOB killed \n\n"
	211	fi' exit
	212
	213
	214	# BEI TERMINAL-BREAK:
	215	trap '[[ -f $job_to_send ]] && rm $job_to_send
	216	printf "\n\n+++ SUBJOB killed \n\n"
	217	exit
	218	' 2
	219
	220
	221
	222
	223	# LOKALEN HOSTNAMEN FESTSTELLEN
	224	local_host=$(hostname)
	225
	226
	227
	228	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
	229	# UEBERHAUPT ZULAESSIG IST
[717]	230	# Note: One of the entries for "lck" or "lckordi" always has to be
	231	# comment out, because the hostname (node*) is same for both machines
[1]	232	case $local_host in
	233	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[102]	234	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	235	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[400]	236	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
	237	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
	238	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
	239	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
	240	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
[291]	241	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[749]	242	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[12]	243	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[372]	244	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[693]	245	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
[108]	246	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
[1]	247	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	248	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[108]	249	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
[367]	250	(node*) local_addres=165.132.26.61 local_host=lck;;
[717]	251	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
[693]	252	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
[101]	253	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[1]	254	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	255	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	256	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	257	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[892]	258	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
[400]	259	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
	260	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
	261	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
	262	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
[739]	263	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
[292]	264	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
[210]	265	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
[440]	266	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
[749]	267	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
[5]	268	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
[407]	269	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
[5]	270	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
[1040]	271	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
[5]	272	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[437]	273	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
[892]	274	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
[1]	275	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	276	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	277	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	278	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[789]	279	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
[5]	280	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[622]	281	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
[1]	282	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
[799]	283	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
[1]	284	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
[259]	285	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
[181]	286	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
[196]	287	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
[635]	288	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
[1021]	289	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
[1]	290	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
[622]	291	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
[1]	292	(*) printf "\n +++ \"$local_host\" unknown";
	293	printf "\n please inform S. Raasch!";
	294	locat=parameter; exit;;
	295	esac
	296
	297
	298
	299	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	300	remote_host=$local_host
	301
	302
	303
	304
	305	# PROZEDUROPTIONEN EINLESEN
[352]	306	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
[1]	307	do
	308	case $option in
	309	(c) job_catalog=$OPTARG;;
	310	(d) delete_dayfile=true;;
	311	(D) no_submit=true;;
[352]	312	(e) email_notification=$OPTARG;;
[125]	313	(g) group_number=$OPTARG;;
[1]	314	(h) remote_host=$OPTARG;;
	315	(m) memory=$OPTARG;;
	316	(n) job_name=$OPTARG;;
	317	(N) node_usage=$OPTARG;;
	318	(O) threads_per_task=$OPTARG;;
	319	(q) no_default_queue=$OPTARG;;
	320	(t) cputime=$OPTARG;;
	321	(T) tasks_per_node=$OPTARG;;
	322	(u) remote_user=$OPTARG;;
	323	(v) verify=false;;
	324	(X) numprocs=$OPTARG;;
	325	(\?) printf "\n +++ Option $OPTARG unknown \n";
	326	locat=option; exit;;
	327	esac
	328	done
	329
	330
	331	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	332	shift OPTIND-1; file_to_send=$1
	333
	334
	335	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	336	if [ "$1" = "?" ]
	337	then
	338	(printf "\n *** subjob can be called as follows:\n"
	339	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	340	printf "\n Description of available options:\n"
	341	printf "\n Option Description Default-Value"
	342	printf "\n -c job-input- and output-catalog ~/job_queue"
	343	printf "\n -d no job-protocol will be created ---"
	344	printf "\n -D only the job-file will be created ---"
	345	printf "\n -h execution host, available hosts: $remote_host"
[693]	346	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
[1040]	347	printf "\n lckiaps, lctit, nech, necriam, unics"
[1]	348	printf "\n -m memory demand per process in MByte ---"
	349	printf "\n -n jobname <jobdatei>"
	350	printf "\n -O threads per task (for OpenMP usage) 1"
	351	printf "\n -q job-queue to be used default"
	352	printf "\n -t allowed cpu-time in seconds ---"
	353	printf "\n -T tasks per node (on parallel hosts) ---"
	354	printf "\n -u username on execution host from .netrc"
	355	printf "\n -v no prompt for confirmation ---"
	356	printf "\n -X # of processors (on parallel hosts) 1"
	357	printf "\n "
	358	printf "\n The only possible positional parameter is <jobfile>:"
	359	printf "\n The complete NQS-job must be provided here."
	360	printf "\n <jobfile>=? creates this outline\n\n") \| more
	361	exit
	362	fi
	363
	364
	365
	366	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	367	if [[ "$file_to_send" = "" ]]
	368	then
	369	printf "\n +++ job-file missing"
	370	locat=parameter; exit
	371	else
	372	if [[ -f $file_to_send ]]
	373	then
	374	true
	375	else
	376	printf "\n +++ job-file: "
	377	printf "\n $file_to_send"
	378	printf "\n does not exist"
	379	locat=parameter; exit
	380	fi
	381	fi
	382
	383
	384
	385	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	386	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	387	if [[ $job_name = none ]]
	388	then
	389	job_name=$file_to_send
	390	fi
	391	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	392	then
	393	printf "\n +++ job-file name: "
	394	printf "\n $job_name"
	395	printf "\n must not contain \"/\"-characters"
	396	locat=parameter; exit
	397	fi
	398
	399
	400
	401
	402	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	403	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	404	if [[ $remote_host = none ]]
	405	then
	406	printf "\n +++ host missing"
	407	locat=option; exit
	408	else
	409	case $remote_host in
[251]	410	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[312]	411	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[693]	412	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[622]	413	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
[251]	414	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	415	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	416	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
[440]	417	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
[540]	418	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
	419	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
[367]	420	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[1040]	421	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
[717]	422	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
[1021]	423	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
[635]	424	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
[892]	425	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
[437]	426	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
[693]	427	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
[251]	428	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	429	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
	430	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	431	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	432	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	433	locat=parameter; exit;;
[1]	434	esac
	435	fi
	436
	437
	438	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	439	if [[ $no_default_queue != none ]]
	440	then
	441	error=false
	442	ndq=$no_default_queue
	443	case $remote_host in
	444	(ibm) case $ndq in
	445	(p690_express\|p690_standard\|p690_long) error=false;;
	446	(*) error=true;;
	447	esac;;
	448	(ibmh) case $ndq in
[312]	449	(no_class) error=false;;
[1]	450	(*) error=true;;
	451	esac;;
[693]	452	(ibmkisti) case $ndq in
	453	(class.32plus\|class.1-2\|class.2-32) error=false;;
	454	(*) error=true;;
	455	esac;;
[622]	456	(ibmku) case $ndq in
	457	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	458	(*) error=true;;
	459	esac;;
[1]	460	(ibms) case $ndq in
	461	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	462	(*) error=true;;
	463	esac;;
	464	(ibmy) case $ndq in
	465	(parallel) error=false;;
	466	(*) error=true;;
	467	esac;;
[1040]	468	(lckiaps) case $ndq in
	469	(express\|normal) error=false;;
	470	(*) error=true;;
	471	esac;;
[440]	472	(lckyoto) case $ndq in
	473	(eh\|ph) error=false;;
	474	(*) error=true;;
	475	esac;;
[191]	476	(lcsgib\|lcsgih) case $ndq in
[416]	477	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
[191]	478	(*) error=true;;
	479	esac;;
[1]	480	(lctit) case $ndq in
[635]	481	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	482	(*) error=true;;
	483	esac;;
	484	(t3eb) case $ndq in
	485	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	486	(*) error=true;;
	487	esac;;
[251]	488	(necriam) case $ndq in
[253]	489	(SP\|SS\|P6) error=false;;
[251]	490	(*) error=true;;
	491	esac;;
[1]	492	(t3eh) case $ndq in
	493	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	494	(*) error=true;;
	495	esac;;
	496	(t3ej2\|t3ej5) case $ndq in
	497	(low\|normal\|high) error=false;;
	498	(*) error=true;;
	499	esac;;
	500	(t3es) case $ndq in
	501	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	502	(*) error=true;;
	503	esac;;
	504	(unics) case $ndq in
	505	(unics\|ht) error=false;;
	506	(*) error=true;;
	507	esac;;
	508	esac
	509	if [[ $error = true ]]
	510	then
	511	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	512	locat=parameter; exit
	513	else
	514	queue=$no_default_queue
	515	fi
	516	fi
	517
	518
	519
	520	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	521	# AUFTEILEN
	522	done=false
	523	while [[ $done = false ]]
	524	do
	525	if (( $cputime <= 0 ))
	526	then
	527	printf "\n +++ wrong cpu-time or cpu-time missing"
	528	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	529	printf "\n >>> "
	530	read cputime 1>/dev/null 2>&1
	531	else
	532	done=true
	533	fi
	534	done
	535	if [[ $remote_host = nech ]]
	536	then
	537	if (( tasks_per_node != 0 ))
	538	then
	539	(( cputime = cputime * tasks_per_node ))
	540	elif [[ $numprocs != 0 ]]
	541	then
	542	(( cputime = cputime * numprocs ))
	543	fi
	544	fi
	545	(( stunden = cputime / 3600 ))
	546	(( resttime = cputime - stunden * 3600 ))
	547	(( minuten = resttime / 60 ))
	548	(( sekunden = resttime - minuten * 60 ))
	549	timestring=${stunden}:${minuten}:${sekunden}
	550
	551
	552
	553	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	554	done=false
	555	while [[ $done = false ]]
	556	do
	557	if (( memory <= 0 ))
	558	then
	559	printf "\n +++ wrong memory demand or memory demand missing"
	560	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	561	printf "\n >>> "
	562	read memory 1>/dev/null 2>&1
	563	else
	564	done=true
	565	fi
	566	done
	567
[251]	568	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
[1]	569	then
	570	if (( tasks_per_node != 0 ))
	571	then
	572	(( Memory = memory * tasks_per_node / 1000 ))
	573	elif [[ $numprocs != 0 ]]
	574	then
	575	(( Memory = memory * numprocs / 1000 ))
	576	else
	577	(( Memory = memory / 1000 ))
	578	fi
[635]	579	elif [[ $remote_host = lctit ]]
	580	then
	581	(( Memory = memory * tasks_per_node / 1000 ))
[1]	582	fi
	583
	584
	585	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	586	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	587	then
	588	(( memory = memory * threads_per_task ))
	589	fi
	590
	591
	592	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	593	if (( tasks_per_node != 0 ))
	594	then
[622]	595	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
[1]	596	fi
	597
[622]	598	# Calculate number of processes per node
	599	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	600
[696]	601	# Calculate number of MPI tasks
	602	(( mpi_tasks = numprocs / threads_per_task ))
[1]	603
[696]	604
[1]	605	# HEADER-AUSGABE
	606	if [[ $verify = true ]]
	607	then
	608	printf "\n\n"
	609	printf "#--------------------------------------------------------------# \n"
	610	spalte1=SUBJOB;spalte2=$(date)
	611	printf "\| $spalte1$spalte2 \| \n"
	612	printf "\| \| \n"
	613	printf "\| values of parameters/options: \| \n"
	614	spalte1=local_host$punkte; spalte2=$punkte$local_host
	615	printf "\| $spalte1$spalte2 \| \n"
	616	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	617	printf "\| $spalte1$spalte2 \| \n"
	618	spalte1=queue$punkte; spalte2=$punkte$queue
	619	printf "\| $spalte1$spalte2 \| \n"
	620	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	621	printf "\| $spalte1$spalte2 \| \n"
	622	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	623	printf "\| $spalte1$spalte2 \| \n"
	624	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	625	printf "\| $spalte1$spalte2 \| \n"
	626	printf "#--------------------------------------------------------------# \n\n"
	627
	628
	629	# KONTROLLABFRAGE, OB ALLES O.K.
	630	antwort="dummy"
	631	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	632	do
	633	read antwort?" >>> continue (y/n) ? "
	634	done
	635	if [[ $antwort = n \|\| $antwort = N ]]
	636	then
	637	locat=verify; exit
	638	fi
	639	printf "\n"
	640	fi
	641
	642
	643
	644	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	645	kennung=$RANDOM
	646	job_on_remhost=${job_name}_${kennung}_$local_host
	647	job_to_send=job_to_send_$kennung
	648	if [[ $delete_dayfile = false ]]
	649	then
	650	remote_dayfile=${local_host}_${job_name}_result_$kennung
	651	local_dayfile=${remote_host}_${job_name}
	652	else
	653	remote_dayfile=/dev/null
	654	fi
	655
	656
	657
[622]	658	# Generate the batch job scripts (qsub/msub/LoadLeveler)
[1]	659	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	660	then
	661
[622]	662	# General LoadLeveler settings
	663	execute_in_shell="#!/bin/ksh"
	664	use_shell="# @ shell = /bin/ksh"
	665	consumable_memory="ConsumableMemory($memory mb)"
	666	class="# @ class = $queue"
	667	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	668	network_to_use="# @ network.mpi = sn_all,shared,us"
	669	data_limit="# @ data_limit = 1.76gb"
	670	image_size="# @ image_size = 50"
[693]	671	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	672
[693]	673	if [[ $email_notification = none ]]
	674	then
	675	notify_user=""
	676	else
	677	notify_user="# @ notify_user = $email_notification"
	678	if [[ $delete_dayfile = true ]]
	679	then
	680	notification='# @ notification = never'
	681	fi
	682	fi
[622]	683
[312]	684	if [[ $remote_host = ibmh ]]
[1]	685	then
[312]	686	data_limit=""
	687	network_to_use=""
	688	class=""
	689	environment=""
[814]	690	rset="# @ rset = RSET_MCM_AFFINITY"
	691	task_affinity="# @ task_affinity = core(1)"
[693]	692	elif [[ $remote_host = ibmkisti ]]
	693	then
	694	network_to_use="# @ network.MPI = sn_all,shared,US"
	695	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	696	if [[ $threads_per_task = 1 ]]
	697	then
	698	rset="# @ rset = RSET_MCM_AFFINITY"
	699	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	700	fi
[693]	701	environment=""
	702	use_shell=""
	703	data_limit=""
	704	image_size=""
[622]	705	elif [[ $remote_host = ibmku ]]
	706	then
	707	execute_in_shell="#!/usr/bin/ksh"
	708	use_shell="# @ shell = /usr/bin/ksh"
	709	consumable_memory=""
	710	environment=""
	711	network_to_use="# @ network.mpi = sn_all,shared,us"
	712	data_limit=""
	713	image_size=""
	714	elif [[ $remote_host = ibms ]]
	715	then
	716	network_to_use="# @ network.mpi = csss,shared,us"
	717	elif [[ $remote_host = ibmy ]]
	718	then
	719	consumable_memory=""
	720	network_to_use=""
[1]	721	fi
	722
	723	cat > $job_to_send << %%END%%
[622]	724	$execute_in_shell
	725	$use_shell
[1]	726
	727	# @ job_type = parallel
[693]	728	# @ job_name = $job_name
[1]	729	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	730	# @ output = $remote_dayfile
	731	# @ error = $remote_dayfile
[693]	732	$wall_clock_limit
[622]	733	$image_size
[312]	734	$class
	735	$environment
[1]	736	$network_to_use
	737	$data_limit
[693]	738	$rset
	739	$mcm_affinity_options
[814]	740	$task_affinity
[1]	741	$notification
[693]	742	$notify_user
[1]	743
	744	%%END%%
	745
	746	if (( nodes > 0 ))
	747	then
	748
[693]	749	if [[ $remote_host != ibmkisti ]]
	750	then
	751
	752	cat >> $job_to_send << %%END%%
[1]	753	# @ node = $nodes
[622]	754	# @ tasks_per_node = $processes_per_node
[1]	755	# @ node_usage = $node_usage
	756	# @ queue
	757
	758	%%END%%
	759
[693]	760	else
	761
	762	cat >> $job_to_send << %%END%%
[696]	763	# @ total_tasks = $mpi_tasks
[693]	764	# @ blocking = unlimited
	765	# @ queue
	766
	767	%%END%%
	768
	769	fi
	770
[1]	771	else
	772
	773	if [[ $remote_host != ibmy ]]
	774	then
	775
	776	cat >> $job_to_send << %%END%%
	777	# @ blocking = unlimited
	778	# @ total_tasks = $numprocs
	779	# @ node_usage = $node_usage
	780	# @ queue
	781
	782	%%END%%
	783
	784	else
	785
	786	cat >> $job_to_send << %%END%%
	787	# @ node = 1
	788	# @ total_tasks = $numprocs
	789	# @ queue
	790
	791	%%END%%
	792
	793	fi
	794
	795	fi
	796
[696]	797	# workaround because of silly job filter on ibmkisti
	798	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	799	then
	800	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	801	fi
	802
[1]	803	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	804	then
	805
	806	cat > $job_to_send << %%END%%
	807	#!/bin/ksh
	808
	809	# @ job_type = serial
	810	# @ node_usage = $node_usage
	811	# @ job_name = palm
	812	# @ wall_clock_limit = ${timestring},$timestring
	813	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	814	# @ output = $remote_dayfile
	815	# @ error = $remote_dayfile
[312]	816	$class
[1]	817	$notification
	818
	819	# @ queue
	820
	821	%%END%%
	822
[108]	823	elif [[ $remote_host = lcfimm ]]
	824	then
	825
	826	if [[ $numprocs != 0 ]]
	827	then
	828	cat > $job_to_send << %%END%%
	829	#!/bin/ksh
	830	#PBS -N $job_name
[552]	831	#PBS -A $project_account
[108]	832	#PBS -l walltime=$timestring
[622]	833	#PBS -l nodes=${nodes}:ppn=$processes_per_node
[108]	834	#PBS -l pmem=${memory}mb
	835	#PBS -m abe
	836	#PBS -o $remote_dayfile
	837	#PBS -j oe
	838	mpd &
	839
	840	%%END%%
	841
[168]	842	else
	843	cat > $job_to_send << %%END%%
	844	#!/bin/ksh
	845	#PBS -N $job_name
[552]	846	#PBS -A $project_account
[168]	847	#PBS -l walltime=$timestring
	848	#PBS -l ncpus=1
	849	#PBS -l pmem=${memory}mb
	850	#PBS -m abe
	851	#PBS -o $remote_dayfile
	852	#PBS -j oe
	853
	854	%%END%%
	855
	856	fi
	857
[1021]	858	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
[368]	859	then
	860
	861	if [[ $numprocs != 0 ]]
	862	then
	863	cat > $job_to_send << %%END%%
	864	#!/bin/ksh
	865	#PBS -N $job_name
	866	#PBS -l walltime=$timestring
	867	#PBS -l ncpus=$numprocs
	868	#PBS -l pmem=${memory}mb
	869	#PBS -o $remote_dayfile
[1021]	870	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[368]	871	#PBS -j oe
	872
	873	mpd &
	874
	875	%%END%%
	876
	877	else
	878	cat > $job_to_send << %%END%%
	879	#!/bin/ksh
	880	#PBS -N $job_name
	881	#PBS -l walltime=$timestring
	882	#PBS -l ncpus=1
	883	#PBS -l pmem=${memory}mb
	884	#PBS -o $remote_dayfile
	885	#PBS -j oe
	886
	887	%%END%%
	888
	889	fi
	890
[1040]	891	elif [[ $remote_host = lckiaps ]]
	892	then
	893
	894	if [[ $numprocs != 0 ]]
	895	then
	896	cat > $job_to_send << %%END%%
	897	#!/bin/ksh
	898	#PBS -N $job_name
	899	#PBS -l walltime=$timestring
	900	#PBS -l select=1:ncpus=$numprocs
	901	#PBS -l pmem=${memory}mb
	902	#PBS -q $queue
	903	#PBS -o $remote_dayfile
	904	#PBS -j oe
	905	#PBS -V
	906
	907	%%END%%
	908
	909	else
	910	cat > $job_to_send << %%END%%
	911	#!/bin/ksh
	912	#PBS -N $job_name
	913	#PBS -l walltime=$timestring
	914	#PBS -l ncpus=1
	915	#PBS -l pmem=${memory}mb
	916	#PBS -o $remote_dayfile
	917	#PBS -j oe
	918
	919	%%END%%
	920
	921	fi
	922
[693]	923	elif [[ $remote_host = lcyon ]]
	924	then
	925
	926	if [[ $numprocs != 0 ]]
	927	then
	928	cat > $job_to_send << %%END%%
	929	#!/bin/ksh
	930	#PBS -N $job_name
	931	#PBS -l walltime=$timestring
	932	#PBS -l ncpus=$numprocs
	933	#PBS -l pmem=${memory}mb
	934	#PBS -o $remote_dayfile
	935	#PBS -j oe
	936
	937	%%END%%
	938
	939	else
	940	cat > $job_to_send << %%END%%
	941	#!/bin/ksh
	942	#PBS -N $job_name
	943	#PBS -l walltime=$timestring
	944	#PBS -l ncpus=1
	945	#PBS -l pmem=${memory}mb
	946	#PBS -o $remote_dayfile
	947	#PBS -j oe
	948
	949	%%END%%
	950
	951	fi
	952
[181]	953	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
[179]	954	then
	955
[376]	956	if [[ "$sgi_feature" != "" ]]
	957	then
	958	feature_directive="#PBS -l feature=$sgi_feature"
	959	else
	960	feature_directive=""
	961	fi
[799]	962
[416]	963	if [[ $queue = dataq \|\| $queue = permq ]]
[399]	964	then
	965	feature_directive="#PBS -l feature=data"
	966	fi
[376]	967
[789]	968	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
[191]	969	then
	970	queue_directive="#PBS -q $queue"
	971	else
	972	queue_directive=""
	973	fi
	974
[352]	975	if [[ $email_notification = none ]]
	976	then
	977	email_directive=""
	978	else
	979	email_directive="#PBS -M $email_notification"
	980	fi
	981
[179]	982	if [[ $numprocs != 0 ]]
	983	then
	984	cat > $job_to_send << %%END%%
[181]	985	#!/bin/bash
[179]	986	#PBS -N $job_name
	987	#PBS -l walltime=$timestring
[622]	988	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[181]	989	#PBS -l naccesspolicy=$node_usage
[179]	990	#PBS -o $remote_dayfile
	991	#PBS -j oe
[376]	992	$feature_directive
[191]	993	$queue_directive
[352]	994	$email_directive
[179]	995
[622]	996	eval \`/sw/swdist/bin/modulesinit\`
	997	#. /usr/share/modules/init/bash
[892]	998	$init_cmds
[493]	999	$module_calls
[374]	1000
[181]	1001	echo ld_library_path=\$LD_LIBRARY_PATH
[179]	1002
	1003	%%END%%
	1004
	1005	else
	1006	cat > $job_to_send << %%END%%
	1007	#PBS -S /bin/bash
	1008	#PBS -N $job_name
	1009	#PBS -l walltime=$timestring
	1010	#PBS -l ncpus=1
	1011	#PBS -o $remote_dayfile
[475]	1012	#PBS -j oe
[400]	1013	$queue_directive
	1014	$email_directive
[179]	1015
[622]	1016	eval \`/sw/swdist/bin/modulesinit\`
	1017	#. /usr/share/modules/init/bash
[892]	1018	$init_cmds
[493]	1019	$module_calls
[475]	1020
[179]	1021	%%END%%
	1022
	1023	fi
	1024
[892]	1025	elif [[ $remote_host = lcxe6 ]]
[164]	1026	then
	1027
	1028	if [[ $numprocs != 0 ]]
	1029	then
	1030	cat > $job_to_send << %%END%%
[799]	1031	#!/bin/ksh
[164]	1032	#PBS -S /bin/ksh
	1033	#PBS -N $job_name
[552]	1034	#PBS -A $project_account
[206]	1035	#PBS -j oe
[164]	1036	#PBS -l walltime=$timestring
	1037	#PBS -l mppwidth=${numprocs}
[622]	1038	#PBS -l mppnppn=${processes_per_node}
[164]	1039	#PBS -m abe
	1040	#PBS -o $remote_dayfile
[492]	1041	$email_directive
[164]	1042
[892]	1043	$init_cmds
[493]	1044	$module_calls
[343]	1045
[164]	1046	%%END%%
	1047
[108]	1048	else
	1049	cat > $job_to_send << %%END%%
	1050	#!/bin/ksh
[168]	1051	#PBS -S /bin/ksh
[108]	1052	#PBS -N $job_name
[552]	1053	#PBS -A $project_account
[206]	1054	#PBS -j oe
[108]	1055	#PBS -l walltime=$timestring
	1056	#PBS -l ncpus=1
	1057	#PBS -l pmem=${memory}mb
	1058	#PBS -m abe
[492]	1059	$email_directive
[108]	1060	#PBS -o $remote_dayfile
	1061
[892]	1062	$init_cmds
[493]	1063	$module_calls
[343]	1064
[108]	1065	%%END%%
	1066
	1067	fi
	1068
[440]	1069	elif [[ $remote_host = lckyoto ]]
[437]	1070	then
	1071
[440]	1072	cat > $job_to_send << %%END%%
[799]	1073	#!/bin/ksh
[440]	1074	# @\$-o $remote_dayfile
	1075	# @\$-eo -oi
	1076	# @\$-lP 16
[799]	1077	# @\$-lp 1
[440]	1078	# @\$-lm 28gb -llm unlimited -ls unlimited
	1079	# @\$-q $queue
	1080	# @\$-Pvn abs_pack
	1081	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	1082	#. /thin/local/etc/setprofile/intel-11.0.sh
	1083	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1084	. ~/.myprofile
	1085	#. /home2/t/t51254/palm/current_version/myprofile
	1086	#. /thin/apps/pgi/mpi.sh
	1087	#
	1088	env
	1089	#
	1090	set -x
	1091
	1092	%%END%%
	1093
	1094	elif [[ $remote_host = lcxt5m ]]
	1095	then
	1096
[437]	1097	if [[ $numprocs != 0 ]]
	1098	then
	1099	cat > $job_to_send << %%END%%
[799]	1100	#!/bin/ksh
[437]	1101	#PBS -S /bin/ksh
	1102	#PBS -N $job_name
	1103	#PBS -j oe
	1104	#PBS -l walltime=$timestring
	1105	#PBS -l mppwidth=${numprocs}
[622]	1106	#PBS -l mppnppn=${processes_per_node}
[437]	1107	#PBS -m abe
	1108	#PBS -o $remote_dayfile
	1109
[892]	1110	$init_cmds
[493]	1111	$module_calls
[437]	1112
	1113	%%END%%
	1114
	1115	else
	1116	cat > $job_to_send << %%END%%
	1117	#!/bin/ksh
	1118	#PBS -S /bin/ksh
	1119	#PBS -N $job_name
	1120	#PBS -j oe
	1121	#PBS -l walltime=$timestring
	1122	#PBS -l ncpus=1
	1123	#PBS -l pmem=${memory}mb
	1124	#PBS -m abe
	1125	#PBS -o $remote_dayfile
	1126
[892]	1127	$init_cmds
[493]	1128	$module_calls
[437]	1129
	1130	%%END%%
	1131
	1132	fi
	1133
[1]	1134	elif [[ $remote_host = nech ]]
	1135	then
	1136
	1137	if (( nodes > 1 ))
	1138	then
	1139	cat > $job_to_send << %%END%%
	1140	#!/bin/ksh
[622]	1141	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1142	#PBS -l ${qsubmem}=${Memory}gb
	1143	#PBS -b $nodes
	1144	#PBS -o $remote_dayfile
	1145	#PBS -N palm
	1146	#PBS -j o
	1147	#PBS -T mpisx
	1148
	1149	%%END%%
	1150
	1151	elif [[ $numprocs != 0 ]]
	1152	then
	1153	cat > $job_to_send << %%END%%
	1154	#!/bin/ksh
[622]	1155	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1156	#PBS -l ${qsubmem}=${Memory}gb
	1157	#PBS -o $remote_dayfile
	1158	#PBS -N palm
	1159	#PBS -j o
	1160
	1161	%%END%%
	1162
	1163	else
	1164	cat > $job_to_send << %%END%%
	1165	#!/bin/ksh
	1166	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1167	#PBS -o $remote_dayfile
	1168	#PBS -j o
	1169
	1170	%%END%%
	1171
	1172	fi
	1173
[251]	1174	elif [[ $remote_host = necriam ]]
[35]	1175	then
	1176
	1177	if (( nodes > 1 ))
	1178	then
	1179	cat > $job_to_send << %%END%%
	1180	#!/bin/ksh
	1181	#PBS -b $nodes
	1182	#PBS -o $remote_dayfile
[253]	1183	#PBS -N $job_name
[35]	1184	#PBS -j o
[251]	1185	#PBS -v MPIPROGINV=YES
[35]	1186
	1187	%%END%%
	1188
	1189	elif [[ $numprocs != 0 ]]
	1190	then
	1191	cat > $job_to_send << %%END%%
	1192	#!/bin/ksh
	1193	#PBS -q ${queue}
	1194	#PBS -o $remote_dayfile
[253]	1195	#PBS -N $job_name
[35]	1196	#PBS -j o
[251]	1197	#PBS -v MPIPROGINV=YES
[35]	1198
	1199	%%END%%
	1200
	1201	else
	1202	cat > $job_to_send << %%END%%
	1203	#!/bin/ksh
	1204	#PBS -o $remote_dayfile
	1205	#PBS -j o
	1206
	1207	%%END%%
	1208
	1209	fi
	1210
[1]	1211	elif [[ $remote_host = lctit ]]
	1212	then
	1213	cat > $job_to_send << %%END%%
[635]	1214	#!/bin/ksh
[892]	1215	$init_cmds
[678]	1216	$module_calls
	1217
[1]	1218	%%END%%
	1219
	1220	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
[678]	1221	if [[ $tasks_per_node != $processes_per_node ]]
	1222	then
	1223	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -o $remote_dayfile -j oe -q $queue "
	1224	else
	1225	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -l place=scatter -o $remote_dayfile -j oe -q $queue "
	1226	fi
[1]	1227
	1228	else
	1229
	1230	cat > $job_to_send << %%END%%
	1231	# @\$-q ${queue}
	1232	# @\$-l${qsubtime} $timestring
	1233	# @\$-l${qsubmem} ${memory}mb
	1234	# @\$-o $remote_dayfile
	1235	# @\$-eo
	1236
	1237	%%END%%
	1238
	1239	fi
	1240
	1241
	1242	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	1243	# BEI EXIT VERANLASSEN
	1244	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	1245	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	1246	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1247	then
	1248	echo "set +vx" >> $job_to_send
	1249	echo "trap '" >> $job_to_send
	1250	echo "set +vx" >> $job_to_send
[1040]	1251	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lckiaps ]]
[1]	1252	then
[622]	1253	if [[ $remote_host = ibmh ]]
[1]	1254	then
	1255	return_queue=c1
[693]	1256	elif [[ $remote_host = ibmkisti ]]
	1257	then
	1258	return_queue=class.1-2
[622]	1259	elif [[ $remote_host = ibmku ]]
	1260	then
	1261	return_queue=sdbg2
[1]	1262	elif [[ $remote_host = ibms ]]
	1263	then
	1264	return_queue=p_normal
	1265	elif [[ $remote_host = ibmy ]]
	1266	then
	1267	return_queue=serial
[325]	1268	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
	1269	then
[326]	1270	return_queue=serialq
[251]	1271	elif [[ $remote_host = necriam ]]
[1]	1272	then
[253]	1273	return_queue=SP
[1040]	1274	elif [[ $remote_host = lckiaps ]]
	1275	then
	1276	return_queue=express
[1]	1277	else
	1278	return_queue=unknown
	1279	fi
	1280
	1281	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1282	then
	1283
[622]	1284	if [[ $remote_host = ibmku ]]
	1285	then
	1286	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1287	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1288	else
	1289	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1290	fi
[1]	1291	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	1292	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	1293	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	1294	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	1295	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	1296	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
[312]	1297	if [[ $host != "ibmh" ]]
	1298	then
	1299	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	1300	fi
[1]	1301	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	1302	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	1303
	1304	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	1305	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	1306
	1307	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	1308	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
[622]	1309	if [[ $remote_host = ibmku ]]
	1310	then
	1311	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
	1312	fi
[1]	1313	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	1314
	1315	elif [[ $remote_host = nech ]]
	1316	then
	1317	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1318	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1319	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1320	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1321	echo "#PBS -j o" >> $job_to_send
	1322	echo " " >> $job_to_send
	1323	echo "set -x" >> $job_to_send
	1324	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1325	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1326	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1327	echo "%%END%%" >> $job_to_send
	1328
[251]	1329	elif [[ $remote_host = necriam ]]
[35]	1330	then
	1331	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1332	echo "#PBS -q $return_queue" >> $job_to_send
	1333	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1334	echo "#PBS -j o" >> $job_to_send
	1335	echo " " >> $job_to_send
	1336	echo "set -x" >> $job_to_send
	1337	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1338	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1339	echo "%%END%%" >> $job_to_send
	1340
[181]	1341	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1342	then
	1343	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1344	echo "#!/bin/bash" >> $job_to_send
	1345	echo "#PBS -N job_protocol_transfer" >> $job_to_send
[325]	1346	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[181]	1347	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[366]	1348	echo "#PBS -l feature=data" >> $job_to_send
[181]	1349	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1350	echo "#PBS -j oe" >> $job_to_send
	1351	echo " " >> $job_to_send
	1352	echo ". /usr/share/modules/init/bash" >> $job_to_send
	1353	echo "set -x" >> $job_to_send
	1354	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1355	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1356	echo "%%END%%" >> $job_to_send
	1357
[1]	1358	else
	1359
	1360	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1361	echo "# @\\\$-q $return_queue" >> $job_to_send
	1362	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1363	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1364	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1365	then
	1366	echo "# @\$-l mpp_p=0" >> $job_to_send
	1367	fi
	1368	echo '# @\$-lF 10mb' >> $job_to_send
	1369	# echo '# @\$-o /dev/null' >> $job_to_send
	1370	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1371	echo '# @\\\$-eo' >> $job_to_send
	1372	echo " " >> $job_to_send
	1373	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1374	then
	1375	echo "set +vx" >> $job_to_send
	1376	echo ". .profile" >> $job_to_send
	1377	fi
	1378	echo "set -x" >> $job_to_send
	1379	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	1380	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1381	echo "%%END%%" >> $job_to_send
	1382	fi
	1383
	1384	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1385	then
	1386	echo "llsubmit scpjob.$kennung" >> $job_to_send
[181]	1387	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1388	then
[748]	1389	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1390	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1391	echo "msub scpjob.$kennung" >> $job_to_send
[1]	1392	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1393	then
	1394	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	1395	elif [[ $remote_host = t3es ]]
	1396	then
	1397	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
[1043]	1398	elif [[ $remote_host = lckiaps ]]
	1399	then
	1400	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
	1401	echo "ssh ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
	1402	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
[1]	1403	else
[1040]	1404	echo "$submcom scpjob.$kennung" >> $job_to_send
[1]	1405	fi
[1043]	1406	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
[622]	1407	then
	1408	echo "rm scpjob.$kennung" >> $job_to_send
	1409	fi
[1]	1410	if [[ $remote_host = nech ]]
	1411	then
	1412	echo "cd -" >> $job_to_send
	1413	fi
	1414	else
	1415	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1416	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	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
	1425	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	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
	1434	# remove job file
[622]	1435	if [[ $remote_host = lctit \|\| $remote_host = ibmku ]]
[69]	1436	then
	1437	echo " " >> $job_to_send
	1438	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1439	fi
[1]	1440
	1441
	1442
	1443	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	1444	if [[ -z $remote_user ]]
	1445	then
	1446	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	1447	then
	1448	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	1449	fi
	1450	fi
	1451
	1452
	1453
	1454	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	1455	if [[ $no_submit = false ]]
	1456	then
	1457	if [[ $remote_host != $local_host ]]
	1458	then
	1459	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	1460	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	1461	then
[82]	1462	job_catalog_save=$job_catalog
	1463	job_catalog=job_queue
	1464	elif [[ $remote_host = nech ]]
	1465	then
	1466	job_catalog_save=$job_catalog
	1467	job_catalog=/hpf/b/${remote_user}/job_queue
	1468	fi
	1469	if [[ $local_host = decalpha ]]
	1470	then
	1471	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	1472	# PASSWORT AB
	1473	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	1474	elif [[ $remote_host = nech ]]
	1475	then
	1476	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	1477	# TRANSFERIERT WERDEN
	1478	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1479	else
[82]	1480	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1481	fi
[82]	1482	if [[ $? = 1 ]]
	1483	then
	1484	locat=scp; exit
	1485	fi
	1486	if [[ $remote_host = ibms ]]
	1487	then
	1488	job_catalog=$job_catalog_save
	1489	fi
[1]	1490	[[ $verify = true ]] && printf "\n >>> finished\n"
	1491	else
	1492	eval job_catalog=$job_catalog
	1493	cp $job_to_send ${job_catalog}/$job_on_remhost
	1494	fi
	1495
	1496
	1497
	1498	# NQS- BZW. LOADLEVELER-JOB STARTEN
	1499	if [[ $remote_host != $local_host ]]
	1500	then
	1501	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1502
	1503	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
[1]	1504	then
[635]	1505	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
	1506	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
	1507	elif [[ $remote_host = ibmku ]]
	1508	then
	1509	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1]	1510	else
[635]	1511	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1512	fi
[1]	1513
	1514	[[ $verify = true ]] && printf " >>> o.k.\n"
	1515	else
	1516	cd $job_catalog
[210]	1517	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
[1]	1518	then
	1519	eval $submcom $job_on_remhost
[1021]	1520	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb ]]
[108]	1521	then
[635]	1522	chmod u+x $job_on_remhost
	1523	echo "$submcom $job_on_remhost"
[108]	1524	eval $submcom $job_on_remhost
[1]	1525	elif [[ $local_host = nech ]]
	1526	then
	1527	if [[ $queue = default ]]
	1528	then
[799]	1529	eval $submcom $job_on_remhost
[1]	1530	else
[799]	1531	eval $submcom -q $queue $job_on_remhost
[1]	1532	fi
	1533	else
	1534	qsub $job_on_remhost
	1535	fi
[622]	1536
	1537	# Jobfile must not be deleted on lctit/ibmku!! This will be done
	1538	# only at the end of the job.
	1539	if [[ $local_host != lctit && $local_host != ibmku ]]
	1540	then
	1541	rm $job_on_remhost
	1542	fi
[1]	1543	cd - > /dev/null
	1544	fi
	1545	fi
	1546
	1547
	1548
	1549	# ABSCHLUSSARBEITEN
	1550
	1551	if [[ $no_submit = false ]]
	1552	then
	1553	rm $job_to_send
	1554	fi
	1555	[[ $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 |