Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 784

Last change on this file since 784 was 749, checked in by raasch, 13 years ago
host inferno admitted
Property svn:keywords set to `Id Rev`
File size: 49.7 KB

Rev	Line
[1]	1	#!/bin/ksh
[82]	2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
[169]	3	# $Id: subjob 749 2011-08-21 10:49:37Z raasch $
[1]	4
	5	# Prozedur zum automatischen Generieren von Batch-Jobs, die unter NQS
	6	# laufen sollen und deren Ergebnis (Dayfile) zum Job-generierenden
	7	# Host zurueckgeschickt werden sollen
	8
	9
	10	# letzte Aenderung:
	11	# 29/06/94 - Siggi - Beginn mit Erstellung der Originalversion
	12	# 08/07/94 - Siggi - Originalversion abgeschlossen (Version 1.0)
	13	# 06/02/98 - Siggi - berte validiert
	14	# 27/01/01 - Siggi - ground.yonsei.ac.kr validiert, Jobs zur T3E in Korea
	15	# moeglich
	16	# 08/02/01 - Siggi - alle subjob-Meldungen ins englische uebersetzt
	17	# 25/05/02 - Siggi - Unterstuetzung des LoadLeveler
	18	# 30/05/02 - Siggi - Validierung fuer ibm-Rechner in Seoul (nobel) sowie
	19	# allgemeine Anpassungen fuer ibm-Rechner
	20	# 15/10/02 - Siggi - Neue Default-Jobklasse (p_normal) fuer IBM in Seoul
	21	# Ruecktransfer des Jobprotokolls fuer diese
	22	# Maschine abgeschaltet
	23	# 31/10/02 - Siggi - berni validiert
	24	# 06/11/02 - Siggi - Neue Jobklassen auf ibmb und ibmh
	25	# 08/11/02 - Siggi - quanero validiert
	26	# 11/12/02 - Siggi - Notification fuer Transfer-Jobs abgeschaltet
	27	# 23/01/03 - Siggi - hostname nobel changed to nobela
	28	# 06/02/03 - Siggi - gregale validated
	29	# 12/02/03 - Siggi - orkan and maestro validated
	30	# 21/02/03 - Siggi - all nobel nodes in Seoul validated
	31	# 12/03/03 - Siggi - nec at DKRZ validated
	32	# 13/03/03 - Siggi - new nqs resource variable Memory
	33	# 07/04/03 - Siggi - processor request option -c on nech needs tasks per
	34	# node
	35	# 11/04/03 - Siggi - network on ibms has attribute "shared"
	36	# 31/07/03 - Siggi - nqs2 on nech implemented (provisional: -h nech2)
	37	# cxxl added to ibmh
	38	# 29/08/03 - Siggi - changes in job queues and communication system on
	39	# ibms
	40	# 24/10/03 - Siggi - using alternate hanni address 130.75.4.2
	41	# 30/10/03 - Siggi - nech is not supported any more
	42	# 10/11/03 - Siggi - nech2 renamed to nech
	43	# 20/11/03 - Siggi - submit command on nech changed from qsub.test to qsub
	44	# 29/03/04 - Siggi - ground not supported any more, gfdl3 validated
	45	# 31/03/04 - Siggi - new option -N for node usage
	46	# 12/04/04 - Siggi - scp2 instead of scp used for transfer from decalpha
	47	# due to error in ssh installation (otherwise a prompt
	48	# for the password appears)
	49	# 23/07/04 - Siggi - changes due to the new berni configuration
	50	# (federation switch)
	51	# 01/09/04 - Gerald new job-classes on hanni
	52	# 08/09/04 - Siggi - hanni IP address changed to 130.75.4.10
	53	# 23/11/04 - Siggi - new job class cdata on hanni and berni
	54	# 03/12/04 - Siggi - notification on ibm switched of in case of
	55	# delete_dayfile = true, node usage in cdev set to
	56	# shared
	57	# 16/02/05 - Gerald hababai validated
	58	# 29/03/05 - Micha - new job class channi on hanni
	59	# 11/05/05 - Siggi - ConsumableMemory is now required as resource keyword
	60	# on ibms
	61	# 24/05/05 - Siggi - Default queue on ibms changed from p_normal_1.3 to
	62	# p_normal
	63	# 30/06/05 - Siggi - network changed for queue cdev from "us" to "ip"
	64	# 12/07/05 - Siggi - in network.mpi on ibmh/ibmb "csss" changed to
	65	# "sn_all", new job class cexp
	66	# 08/09/05 - Siggi - IP-address of gfdl3 changed
	67	# 31/10/05 - Siggi - new job class pp on hurricane, serial jobs on
	68	# hurricane (with -X0)
	69	# 01/11/05 - Siggi - missing queue for jobs submitted on nech (for nech)
	70	# added
	71	# 30/12/05 - Siggi - change of IP adresses in subnet 130.75.105
	72	# 09/02/06 - Siggi - ibmy admitted
	73	# 10/02/06 - Siggi - scp2 changed to /bin/scp on decalpha
	74	# 13/04/06 - Siggi - ostria admitted
	75	# 18/04/06 - Siggi - new option -O for OpenMP usage
	76	# 24/05/06 - Siggi - lctit admitted, ftpjob renamed scpjob
	77	# 25/07/06 - Siggi - gfdl5 (ibmy) admitted for submitting jobs
	78	# 27/09/06 - Siggi - breg/hreg extended with berni/hanni
	79	# 25/10/06 - Siggi - data_limit set to 1.76 GByte on hanni and berni
	80	# 28/11/06 - Siggi - levanto admitted
[5]	81	# 13/02/07 - Siggi - hpmuk releated code removed
[35]	82	# 01/03/07 - Siggi - adjustments for RIAM machines gate and NEC-SX8 (n-sx)
[82]	83	# 12/04/07 - Siggi - option -f (filetransfer protocol) removed, scp only
[102]	84	# 27/07/07 - Siggi - autan admitted
[108]	85	# 03/08/07 - Marcus- lcfimm admitted
[113]	86	# 08/10/07 - Siggi - further job classes added for hanni (csoloh...)
[118]	87	# 15/10/07 - Siggi - preliminary adjustments for lctit based on Jin's
	88	# suggestions
[122]	89	# 19/10/07 - Marcus- add new optional argument -g group_number
[125]	90	# 19/10/07 - Siggi - a ";" was missing in the last change done by Marcus
[129]	91	# 30/10/07 - Marcus- further adjustments for queues on lctit
[164]	92	# 15/05/08 - Siggi - adjustments for lcxt4 (Bergen Center for Computational
	93	# Science)
[179]	94	# 14/07/08 - Siggi - adjustments for lcsgih
[205]	95	# 23/09/08 - Gerald- paesano admitted
[206]	96	# 02/10/08 - Siggi - PBS adjustments for lcxt4
[251]	97	# 02/03/09 - Siggi - Adjustments for new NEC-SX9 at RIAM
[292]	98	# 16/04/09 - Marcus- Adjustments for lcsgib and lcsgih
[312]	99	# 21/04/09 - Siggi - adjustments for new IBM at DKRZ, which is now ibmh
[325]	100	# 18/05/09 - Siggi - Settings for serial jobs on lcsgi changed
[343]	101	# 24/06/09 - BjornM- adjustments for lcxt4 (loading modules manually)
[352]	102	# 08/07/09 - Siggi - option -e added (email notification on lcsgih/b)
[366]	103	# 20/07/09 - Siggi - On lcsgi, jobs for returning the job protocol are
	104	# now run on the data nodes (feature=data)
[367]	105	# 25/08/09 - BjornM- adapted for lck
[372]	106	# 26/08/09 - Marcus- caurus admitted; optional qos feature hiprio on lcsgi
[376]	107	# 03/09/09 - Siggi - PBS sgi feature directive only used if explicitly
	108	# set in the config file by the user
[416]	109	# 16/10/09 - Carolin-adjustments for archiving on SGI-ICE of binary files;
	110	# adding special1q
[407]	111	# 01/12/09 - BjornM- re-adjustments for lcxt4, added sno (130.75.105.113)
[416]	112	# 16/10/09 - Carolin-adding permq
[440]	113	# 01/02/10 - Siggi - adapted for lcxt5m and lckyoto (Fujitsu HX600)
[475]	114	# 03/02/10 - Siggi - bug in serial jobs removed
[492]	115	# 26/02/10 - BjornM- re-adjustments for lcxt4 (new modules, email
	116	# notification)
[493]	117	# 01/03/10 - Siggi - loading of modules controlled by environment variable
	118	# module_calls
[551]	119	# 17/08/10 - BjornM- re-adjustments for lcxt4 (location of qsub)
	120	# 25/08/10 - BjornM- account geofysisk replaced by guest for lcxt4
[552]	121	# 25/08/10 - Siggi - new variable project_account in pbs-statements for
	122	# lcxt4
[622]	123	# 08/12/10 - Siggi - initialization of the module command changed for
	124	# SGI-ICE/lcsgi
	125	# adjustments for Kyushu Univ. (lcrte, ibmku)
[635]	126	# 14/12/10 - Siggi - adjustments for new Tsubame system at Tokyo
	127	# institute of technology (lctit)
[678]	128	# 02/02/11 - Siggi - further asjustments for Tsubame concerning openMP
[693]	129	# 06/03/11 - Siggi - adjustments for ibmkisti
[696]	130	# 17/03/11 - Siggi - adjustments for openmp usage on ibmkisti
[717]	131	# 03/04/11 - Micha - added lckordi
[739]	132	# 17/08/11 - Siggi - hicegate0 added
[748]	133	# 18/08/11 - Siggi - workaround on lcsgi in order to avoid appends to file
	134	# last_job_transfer_protocol
[749]	135	# 21/08/11 - Siggi - inferno admitted
[1]	136
[437]	137
[552]	138
[1]	139	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
[352]	140	delete_dayfile=false
	141	email_notification=none
[122]	142	group_number=none
[1]	143	locat=normal
	144	no_default_queue=none
	145	no_submit=false
	146	job_catalog="~/job_queue"
	147	job_name=none
	148	local_user=$LOGNAME
	149	node_usage=shared
[475]	150	numprocs=0
[1]	151	punkte="..........................................................."
	152	submcom=qsub
	153	queue=default
	154	remote_host=none
	155	remote_user=""
	156	verify=true
	157
	158	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	159	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	160	typeset -L20 spalte1
	161	typeset -R40 spalte2
	162	typeset -L60 spalte3
	163
	164
	165
	166	# FEHLERBEHANDLUNG
	167	# BEI EXIT:
	168	trap 'if [[ $locat != normal ]]
	169	then
	170	case $locat in
	171	(option) printf "\n --> available optios can be displayed"
	172	printf " by typing:"
	173	printf "\n \"subjob ?\" \n";;
	174	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	175	(*) printf "\n +++ unknown error"
	176	printf "\n please inform S. Raasch!\n"
	177	esac
	178	[[ -f $job_to_send ]] && rm $job_to_send
	179	printf "\n\n+++ SUBJOB killed \n\n"
	180	fi' exit
	181
	182
	183	# BEI TERMINAL-BREAK:
	184	trap '[[ -f $job_to_send ]] && rm $job_to_send
	185	printf "\n\n+++ SUBJOB killed \n\n"
	186	exit
	187	' 2
	188
	189
	190
	191
	192	# LOKALEN HOSTNAMEN FESTSTELLEN
	193	local_host=$(hostname)
	194
	195
	196
	197	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
	198	# UEBERHAUPT ZULAESSIG IST
[717]	199	# Note: One of the entries for "lck" or "lckordi" always has to be
	200	# comment out, because the hostname (node*) is same for both machines
[1]	201	case $local_host in
	202	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[102]	203	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	204	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[400]	205	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
	206	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
	207	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
	208	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
	209	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
[291]	210	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[749]	211	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[12]	212	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[372]	213	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[693]	214	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
[108]	215	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
[1]	216	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	217	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[108]	218	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
[367]	219	(node*) local_addres=165.132.26.61 local_host=lck;;
[717]	220	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
[693]	221	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
[101]	222	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[1]	223	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	224	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	225	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	226	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[164]	227	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
[400]	228	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
	229	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
	230	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
	231	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
[739]	232	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
[292]	233	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
[210]	234	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
[440]	235	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
[749]	236	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
[5]	237	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
[407]	238	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
[5]	239	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
	240	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[437]	241	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
[164]	242	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
[1]	243	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	244	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	245	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	246	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[5]	247	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[622]	248	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
[1]	249	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
	250	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
[259]	251	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
[181]	252	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
[196]	253	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
[635]	254	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
[1]	255	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
[622]	256	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
[1]	257	(*) printf "\n +++ \"$local_host\" unknown";
	258	printf "\n please inform S. Raasch!";
	259	locat=parameter; exit;;
	260	esac
	261
	262
	263
	264	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	265	remote_host=$local_host
	266
	267
	268
	269
	270	# PROZEDUROPTIONEN EINLESEN
[352]	271	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
[1]	272	do
	273	case $option in
	274	(c) job_catalog=$OPTARG;;
	275	(d) delete_dayfile=true;;
	276	(D) no_submit=true;;
[352]	277	(e) email_notification=$OPTARG;;
[125]	278	(g) group_number=$OPTARG;;
[1]	279	(h) remote_host=$OPTARG;;
	280	(m) memory=$OPTARG;;
	281	(n) job_name=$OPTARG;;
	282	(N) node_usage=$OPTARG;;
	283	(O) threads_per_task=$OPTARG;;
	284	(q) no_default_queue=$OPTARG;;
	285	(t) cputime=$OPTARG;;
	286	(T) tasks_per_node=$OPTARG;;
	287	(u) remote_user=$OPTARG;;
	288	(v) verify=false;;
	289	(X) numprocs=$OPTARG;;
	290	(\?) printf "\n +++ Option $OPTARG unknown \n";
	291	locat=option; exit;;
	292	esac
	293	done
	294
	295
	296	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	297	shift OPTIND-1; file_to_send=$1
	298
	299
	300	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	301	if [ "$1" = "?" ]
	302	then
	303	(printf "\n *** subjob can be called as follows:\n"
	304	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	305	printf "\n Description of available options:\n"
	306	printf "\n Option Description Default-Value"
	307	printf "\n -c job-input- and output-catalog ~/job_queue"
	308	printf "\n -d no job-protocol will be created ---"
	309	printf "\n -D only the job-file will be created ---"
	310	printf "\n -h execution host, available hosts: $remote_host"
[693]	311	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
[251]	312	printf "\n lctit, nech, necriam, unics"
[1]	313	printf "\n -m memory demand per process in MByte ---"
	314	printf "\n -n jobname <jobdatei>"
	315	printf "\n -O threads per task (for OpenMP usage) 1"
	316	printf "\n -q job-queue to be used default"
	317	printf "\n -t allowed cpu-time in seconds ---"
	318	printf "\n -T tasks per node (on parallel hosts) ---"
	319	printf "\n -u username on execution host from .netrc"
	320	printf "\n -v no prompt for confirmation ---"
	321	printf "\n -X # of processors (on parallel hosts) 1"
	322	printf "\n "
	323	printf "\n The only possible positional parameter is <jobfile>:"
	324	printf "\n The complete NQS-job must be provided here."
	325	printf "\n <jobfile>=? creates this outline\n\n") \| more
	326	exit
	327	fi
	328
	329
	330
	331	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	332	if [[ "$file_to_send" = "" ]]
	333	then
	334	printf "\n +++ job-file missing"
	335	locat=parameter; exit
	336	else
	337	if [[ -f $file_to_send ]]
	338	then
	339	true
	340	else
	341	printf "\n +++ job-file: "
	342	printf "\n $file_to_send"
	343	printf "\n does not exist"
	344	locat=parameter; exit
	345	fi
	346	fi
	347
	348
	349
	350	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	351	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	352	if [[ $job_name = none ]]
	353	then
	354	job_name=$file_to_send
	355	fi
	356	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	357	then
	358	printf "\n +++ job-file name: "
	359	printf "\n $job_name"
	360	printf "\n must not contain \"/\"-characters"
	361	locat=parameter; exit
	362	fi
	363
	364
	365
	366
	367	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	368	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	369	if [[ $remote_host = none ]]
	370	then
	371	printf "\n +++ host missing"
	372	locat=option; exit
	373	else
	374	case $remote_host in
[251]	375	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[312]	376	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[693]	377	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[622]	378	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
[251]	379	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	380	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	381	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
[440]	382	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
[540]	383	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
	384	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
[367]	385	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[717]	386	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
[635]	387	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
[544]	388	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.4.9-snap.201005191035/bin/qsub;;
[437]	389	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
[693]	390	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
[251]	391	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	392	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
	393	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	394	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	395	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	396	locat=parameter; exit;;
[1]	397	esac
	398	fi
	399
	400
	401	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	402	if [[ $no_default_queue != none ]]
	403	then
	404	error=false
	405	ndq=$no_default_queue
	406	case $remote_host in
	407	(ibm) case $ndq in
	408	(p690_express\|p690_standard\|p690_long) error=false;;
	409	(*) error=true;;
	410	esac;;
	411	(ibmh) case $ndq in
[312]	412	(no_class) error=false;;
[1]	413	(*) error=true;;
	414	esac;;
[693]	415	(ibmkisti) case $ndq in
	416	(class.32plus\|class.1-2\|class.2-32) error=false;;
	417	(*) error=true;;
	418	esac;;
[622]	419	(ibmku) case $ndq in
	420	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	421	(*) error=true;;
	422	esac;;
[1]	423	(ibms) case $ndq in
	424	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	425	(*) error=true;;
	426	esac;;
	427	(ibmy) case $ndq in
	428	(parallel) error=false;;
	429	(*) error=true;;
	430	esac;;
[440]	431	(lckyoto) case $ndq in
	432	(eh\|ph) error=false;;
	433	(*) error=true;;
	434	esac;;
[191]	435	(lcsgib\|lcsgih) case $ndq in
[416]	436	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
[191]	437	(*) error=true;;
	438	esac;;
[1]	439	(lctit) case $ndq in
[635]	440	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	441	(*) error=true;;
	442	esac;;
	443	(t3eb) case $ndq in
	444	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	445	(*) error=true;;
	446	esac;;
[251]	447	(necriam) case $ndq in
[253]	448	(SP\|SS\|P6) error=false;;
[251]	449	(*) error=true;;
	450	esac;;
[1]	451	(t3eh) case $ndq in
	452	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	453	(*) error=true;;
	454	esac;;
	455	(t3ej2\|t3ej5) case $ndq in
	456	(low\|normal\|high) error=false;;
	457	(*) error=true;;
	458	esac;;
	459	(t3es) case $ndq in
	460	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	461	(*) error=true;;
	462	esac;;
	463	(unics) case $ndq in
	464	(unics\|ht) error=false;;
	465	(*) error=true;;
	466	esac;;
	467	esac
	468	if [[ $error = true ]]
	469	then
	470	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	471	locat=parameter; exit
	472	else
	473	queue=$no_default_queue
	474	fi
	475	fi
	476
	477
	478
	479	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	480	# AUFTEILEN
	481	done=false
	482	while [[ $done = false ]]
	483	do
	484	if (( $cputime <= 0 ))
	485	then
	486	printf "\n +++ wrong cpu-time or cpu-time missing"
	487	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	488	printf "\n >>> "
	489	read cputime 1>/dev/null 2>&1
	490	else
	491	done=true
	492	fi
	493	done
	494	if [[ $remote_host = nech ]]
	495	then
	496	if (( tasks_per_node != 0 ))
	497	then
	498	(( cputime = cputime * tasks_per_node ))
	499	elif [[ $numprocs != 0 ]]
	500	then
	501	(( cputime = cputime * numprocs ))
	502	fi
	503	fi
	504	(( stunden = cputime / 3600 ))
	505	(( resttime = cputime - stunden * 3600 ))
	506	(( minuten = resttime / 60 ))
	507	(( sekunden = resttime - minuten * 60 ))
	508	timestring=${stunden}:${minuten}:${sekunden}
	509
	510
	511
	512	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	513	done=false
	514	while [[ $done = false ]]
	515	do
	516	if (( memory <= 0 ))
	517	then
	518	printf "\n +++ wrong memory demand or memory demand missing"
	519	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	520	printf "\n >>> "
	521	read memory 1>/dev/null 2>&1
	522	else
	523	done=true
	524	fi
	525	done
	526
[251]	527	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
[1]	528	then
	529	if (( tasks_per_node != 0 ))
	530	then
	531	(( Memory = memory * tasks_per_node / 1000 ))
	532	elif [[ $numprocs != 0 ]]
	533	then
	534	(( Memory = memory * numprocs / 1000 ))
	535	else
	536	(( Memory = memory / 1000 ))
	537	fi
[635]	538	elif [[ $remote_host = lctit ]]
	539	then
	540	(( Memory = memory * tasks_per_node / 1000 ))
[1]	541	fi
	542
	543
	544	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	545	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	546	then
	547	(( memory = memory * threads_per_task ))
	548	fi
	549
	550
	551	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	552	if (( tasks_per_node != 0 ))
	553	then
[622]	554	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
[1]	555	fi
	556
[622]	557	# Calculate number of processes per node
	558	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	559
[696]	560	# Calculate number of MPI tasks
	561	(( mpi_tasks = numprocs / threads_per_task ))
[1]	562
[696]	563
[1]	564	# HEADER-AUSGABE
	565	if [[ $verify = true ]]
	566	then
	567	printf "\n\n"
	568	printf "#--------------------------------------------------------------# \n"
	569	spalte1=SUBJOB;spalte2=$(date)
	570	printf "\| $spalte1$spalte2 \| \n"
	571	printf "\| \| \n"
	572	printf "\| values of parameters/options: \| \n"
	573	spalte1=local_host$punkte; spalte2=$punkte$local_host
	574	printf "\| $spalte1$spalte2 \| \n"
	575	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	576	printf "\| $spalte1$spalte2 \| \n"
	577	spalte1=queue$punkte; spalte2=$punkte$queue
	578	printf "\| $spalte1$spalte2 \| \n"
	579	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	580	printf "\| $spalte1$spalte2 \| \n"
	581	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	582	printf "\| $spalte1$spalte2 \| \n"
	583	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	584	printf "\| $spalte1$spalte2 \| \n"
	585	printf "#--------------------------------------------------------------# \n\n"
	586
	587
	588	# KONTROLLABFRAGE, OB ALLES O.K.
	589	antwort="dummy"
	590	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	591	do
	592	read antwort?" >>> continue (y/n) ? "
	593	done
	594	if [[ $antwort = n \|\| $antwort = N ]]
	595	then
	596	locat=verify; exit
	597	fi
	598	printf "\n"
	599	fi
	600
	601
	602
	603	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	604	kennung=$RANDOM
	605	job_on_remhost=${job_name}_${kennung}_$local_host
	606	job_to_send=job_to_send_$kennung
	607	if [[ $delete_dayfile = false ]]
	608	then
	609	remote_dayfile=${local_host}_${job_name}_result_$kennung
	610	local_dayfile=${remote_host}_${job_name}
	611	else
	612	remote_dayfile=/dev/null
	613	fi
	614
	615
	616
[622]	617	# Generate the batch job scripts (qsub/msub/LoadLeveler)
[1]	618	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	619	then
	620
[622]	621	# General LoadLeveler settings
	622	execute_in_shell="#!/bin/ksh"
	623	use_shell="# @ shell = /bin/ksh"
	624	consumable_memory="ConsumableMemory($memory mb)"
	625	class="# @ class = $queue"
	626	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	627	network_to_use="# @ network.mpi = sn_all,shared,us"
	628	data_limit="# @ data_limit = 1.76gb"
	629	image_size="# @ image_size = 50"
[693]	630	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	631
[693]	632	if [[ $email_notification = none ]]
	633	then
	634	notify_user=""
	635	else
	636	notify_user="# @ notify_user = $email_notification"
	637	if [[ $delete_dayfile = true ]]
	638	then
	639	notification='# @ notification = never'
	640	fi
	641	fi
[622]	642
[312]	643	if [[ $remote_host = ibmh ]]
[1]	644	then
[312]	645	data_limit=""
	646	network_to_use=""
	647	class=""
	648	environment=""
[693]	649	elif [[ $remote_host = ibmkisti ]]
	650	then
	651	network_to_use="# @ network.MPI = sn_all,shared,US"
	652	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	653	if [[ $threads_per_task = 1 ]]
	654	then
	655	rset="# @ rset = RSET_MCM_AFFINITY"
	656	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	657	fi
[693]	658	environment=""
	659	use_shell=""
	660	data_limit=""
	661	image_size=""
[622]	662	elif [[ $remote_host = ibmku ]]
	663	then
	664	execute_in_shell="#!/usr/bin/ksh"
	665	use_shell="# @ shell = /usr/bin/ksh"
	666	consumable_memory=""
	667	environment=""
	668	network_to_use="# @ network.mpi = sn_all,shared,us"
	669	data_limit=""
	670	image_size=""
	671	elif [[ $remote_host = ibms ]]
	672	then
	673	network_to_use="# @ network.mpi = csss,shared,us"
	674	elif [[ $remote_host = ibmy ]]
	675	then
	676	consumable_memory=""
	677	network_to_use=""
[1]	678	fi
	679
	680	cat > $job_to_send << %%END%%
[622]	681	$execute_in_shell
	682	$use_shell
[1]	683
	684	# @ job_type = parallel
[693]	685	# @ job_name = $job_name
[1]	686	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	687	# @ output = $remote_dayfile
	688	# @ error = $remote_dayfile
[693]	689	$wall_clock_limit
[622]	690	$image_size
[312]	691	$class
	692	$environment
[1]	693	$network_to_use
	694	$data_limit
[693]	695	$rset
	696	$mcm_affinity_options
[1]	697	$notification
[693]	698	$notify_user
[1]	699
	700	%%END%%
	701
	702	if (( nodes > 0 ))
	703	then
	704
[693]	705	if [[ $remote_host != ibmkisti ]]
	706	then
	707
	708	cat >> $job_to_send << %%END%%
[1]	709	# @ node = $nodes
[622]	710	# @ tasks_per_node = $processes_per_node
[1]	711	# @ node_usage = $node_usage
	712	# @ queue
	713
	714	%%END%%
	715
[693]	716	else
	717
	718	cat >> $job_to_send << %%END%%
[696]	719	# @ total_tasks = $mpi_tasks
[693]	720	# @ blocking = unlimited
	721	# @ queue
	722
	723	%%END%%
	724
	725	fi
	726
[1]	727	else
	728
	729	if [[ $remote_host != ibmy ]]
	730	then
	731
	732	cat >> $job_to_send << %%END%%
	733	# @ blocking = unlimited
	734	# @ total_tasks = $numprocs
	735	# @ node_usage = $node_usage
	736	# @ queue
	737
	738	%%END%%
	739
	740	else
	741
	742	cat >> $job_to_send << %%END%%
	743	# @ node = 1
	744	# @ total_tasks = $numprocs
	745	# @ queue
	746
	747	%%END%%
	748
	749	fi
	750
	751	fi
	752
[696]	753	# workaround because of silly job filter on ibmkisti
	754	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	755	then
	756	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	757	fi
	758
[1]	759	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	760	then
	761
	762	cat > $job_to_send << %%END%%
	763	#!/bin/ksh
	764
	765	# @ job_type = serial
	766	# @ node_usage = $node_usage
	767	# @ job_name = palm
	768	# @ wall_clock_limit = ${timestring},$timestring
	769	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	770	# @ output = $remote_dayfile
	771	# @ error = $remote_dayfile
[312]	772	$class
[1]	773	$notification
	774
	775	# @ queue
	776
	777	%%END%%
	778
[108]	779	elif [[ $remote_host = lcfimm ]]
	780	then
	781
	782	if [[ $numprocs != 0 ]]
	783	then
	784	cat > $job_to_send << %%END%%
	785	#!/bin/ksh
	786	#PBS -N $job_name
[552]	787	#PBS -A $project_account
[108]	788	#PBS -l walltime=$timestring
[622]	789	#PBS -l nodes=${nodes}:ppn=$processes_per_node
[108]	790	#PBS -l pmem=${memory}mb
	791	#PBS -m abe
	792	#PBS -o $remote_dayfile
	793	#PBS -j oe
	794	mpd &
	795
	796	%%END%%
	797
[168]	798	else
	799	cat > $job_to_send << %%END%%
	800	#!/bin/ksh
	801	#PBS -N $job_name
[552]	802	#PBS -A $project_account
[168]	803	#PBS -l walltime=$timestring
	804	#PBS -l ncpus=1
	805	#PBS -l pmem=${memory}mb
	806	#PBS -m abe
	807	#PBS -o $remote_dayfile
	808	#PBS -j oe
	809
	810	%%END%%
	811
	812	fi
	813
[717]	814	elif [[ $remote_host = lck \|\| $remote_host = lckordi ]]
[368]	815	then
	816
	817	if [[ $numprocs != 0 ]]
	818	then
	819	cat > $job_to_send << %%END%%
	820	#!/bin/ksh
	821	#PBS -N $job_name
	822	#PBS -l walltime=$timestring
	823	#PBS -l ncpus=$numprocs
	824	#PBS -l pmem=${memory}mb
	825	#PBS -o $remote_dayfile
	826	#PBS -j oe
	827
	828	mpd &
	829
	830	%%END%%
	831
	832	else
	833	cat > $job_to_send << %%END%%
	834	#!/bin/ksh
	835	#PBS -N $job_name
	836	#PBS -l walltime=$timestring
	837	#PBS -l ncpus=1
	838	#PBS -l pmem=${memory}mb
	839	#PBS -o $remote_dayfile
	840	#PBS -j oe
	841
	842	%%END%%
	843
	844	fi
	845
[693]	846	elif [[ $remote_host = lcyon ]]
	847	then
	848
	849	if [[ $numprocs != 0 ]]
	850	then
	851	cat > $job_to_send << %%END%%
	852	#!/bin/ksh
	853	#PBS -N $job_name
	854	#PBS -l walltime=$timestring
	855	#PBS -l ncpus=$numprocs
	856	#PBS -l pmem=${memory}mb
	857	#PBS -o $remote_dayfile
	858	#PBS -j oe
	859
	860	%%END%%
	861
	862	else
	863	cat > $job_to_send << %%END%%
	864	#!/bin/ksh
	865	#PBS -N $job_name
	866	#PBS -l walltime=$timestring
	867	#PBS -l ncpus=1
	868	#PBS -l pmem=${memory}mb
	869	#PBS -o $remote_dayfile
	870	#PBS -j oe
	871
	872	%%END%%
	873
	874	fi
	875
[181]	876	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
[179]	877	then
	878
[376]	879	if [[ "$sgi_feature" != "" ]]
	880	then
	881	feature_directive="#PBS -l feature=$sgi_feature"
	882	else
	883	feature_directive=""
	884	fi
[399]	885
[416]	886	if [[ $queue = dataq \|\| $queue = permq ]]
[399]	887	then
	888	feature_directive="#PBS -l feature=data"
	889	fi
[376]	890
[416]	891	if [[ $queue = testq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
[191]	892	then
	893	queue_directive="#PBS -q $queue"
	894	else
	895	queue_directive=""
	896	fi
	897
[352]	898	if [[ $email_notification = none ]]
	899	then
	900	email_directive=""
	901	else
	902	email_directive="#PBS -M $email_notification"
	903	fi
	904
[179]	905	if [[ $numprocs != 0 ]]
	906	then
	907	cat > $job_to_send << %%END%%
[181]	908	#!/bin/bash
[179]	909	#PBS -N $job_name
	910	#PBS -l walltime=$timestring
[622]	911	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[181]	912	#PBS -l naccesspolicy=$node_usage
[179]	913	#PBS -o $remote_dayfile
	914	#PBS -j oe
[376]	915	$feature_directive
[191]	916	$queue_directive
[352]	917	$email_directive
[179]	918
[622]	919	eval \`/sw/swdist/bin/modulesinit\`
	920	#. /usr/share/modules/init/bash
[493]	921	$module_calls
[374]	922
[181]	923	echo ld_library_path=\$LD_LIBRARY_PATH
[179]	924
	925	%%END%%
	926
	927	else
	928	cat > $job_to_send << %%END%%
	929	#PBS -S /bin/bash
	930	#PBS -N $job_name
	931	#PBS -l walltime=$timestring
	932	#PBS -l ncpus=1
	933	#PBS -o $remote_dayfile
[475]	934	#PBS -j oe
[400]	935	$queue_directive
	936	$email_directive
[179]	937
[622]	938	eval \`/sw/swdist/bin/modulesinit\`
	939	#. /usr/share/modules/init/bash
[493]	940	$module_calls
[475]	941
[179]	942	%%END%%
	943
	944	fi
	945
[164]	946	elif [[ $remote_host = lcxt4 ]]
	947	then
	948
	949	if [[ $numprocs != 0 ]]
	950	then
	951	cat > $job_to_send << %%END%%
	952	#!/bin/ksh
	953	#PBS -S /bin/ksh
	954	#PBS -N $job_name
[552]	955	#PBS -A $project_account
[206]	956	#PBS -j oe
[164]	957	#PBS -l walltime=$timestring
	958	#PBS -l mppwidth=${numprocs}
[622]	959	#PBS -l mppnppn=${processes_per_node}
[164]	960	#PBS -m abe
	961	#PBS -o $remote_dayfile
[492]	962	$email_directive
[164]	963
[343]	964	. /opt/modules/default/init/ksh
[493]	965	$module_calls
[343]	966
[164]	967	%%END%%
	968
[108]	969	else
	970	cat > $job_to_send << %%END%%
	971	#!/bin/ksh
[168]	972	#PBS -S /bin/ksh
[108]	973	#PBS -N $job_name
[552]	974	#PBS -A $project_account
[206]	975	#PBS -j oe
[108]	976	#PBS -l walltime=$timestring
	977	#PBS -l ncpus=1
	978	#PBS -l pmem=${memory}mb
	979	#PBS -m abe
[492]	980	$email_directive
[108]	981	#PBS -o $remote_dayfile
	982
[343]	983	. /opt/modules/default/init/ksh
[493]	984	$module_calls
[343]	985
[108]	986	%%END%%
	987
	988	fi
	989
[440]	990	elif [[ $remote_host = lckyoto ]]
[437]	991	then
	992
[440]	993	cat > $job_to_send << %%END%%
	994	#!/bin/ksh
	995	# @\$-o $remote_dayfile
	996	# @\$-eo -oi
	997	# @\$-lP 16
	998	# @\$-lp 1
	999	# @\$-lm 28gb -llm unlimited -ls unlimited
	1000	# @\$-q $queue
	1001	# @\$-Pvn abs_pack
	1002	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	1003	#. /thin/local/etc/setprofile/intel-11.0.sh
	1004	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1005	. ~/.myprofile
	1006	#. /home2/t/t51254/palm/current_version/myprofile
	1007	#. /thin/apps/pgi/mpi.sh
	1008	#
	1009	env
	1010	#
	1011	set -x
	1012
	1013	%%END%%
	1014
	1015	elif [[ $remote_host = lcxt5m ]]
	1016	then
	1017
[437]	1018	if [[ $numprocs != 0 ]]
	1019	then
	1020	cat > $job_to_send << %%END%%
	1021	#!/bin/ksh
	1022	#PBS -S /bin/ksh
	1023	#PBS -N $job_name
	1024	#PBS -j oe
	1025	#PBS -l walltime=$timestring
	1026	#PBS -l mppwidth=${numprocs}
[622]	1027	#PBS -l mppnppn=${processes_per_node}
[437]	1028	#PBS -m abe
	1029	#PBS -o $remote_dayfile
	1030
	1031	. /opt/modules/default/init/ksh
[493]	1032	$module_calls
[437]	1033
	1034	%%END%%
	1035
	1036	else
	1037	cat > $job_to_send << %%END%%
	1038	#!/bin/ksh
	1039	#PBS -S /bin/ksh
	1040	#PBS -N $job_name
	1041	#PBS -j oe
	1042	#PBS -l walltime=$timestring
	1043	#PBS -l ncpus=1
	1044	#PBS -l pmem=${memory}mb
	1045	#PBS -m abe
	1046	#PBS -o $remote_dayfile
	1047
	1048	. /opt/modules/default/init/ksh
[493]	1049	$module_calls
[437]	1050
	1051	%%END%%
	1052
	1053	fi
	1054
[1]	1055	elif [[ $remote_host = nech ]]
	1056	then
	1057
	1058	if (( nodes > 1 ))
	1059	then
	1060	cat > $job_to_send << %%END%%
	1061	#!/bin/ksh
[622]	1062	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1063	#PBS -l ${qsubmem}=${Memory}gb
	1064	#PBS -b $nodes
	1065	#PBS -o $remote_dayfile
	1066	#PBS -N palm
	1067	#PBS -j o
	1068	#PBS -T mpisx
	1069
	1070	%%END%%
	1071
	1072	elif [[ $numprocs != 0 ]]
	1073	then
	1074	cat > $job_to_send << %%END%%
	1075	#!/bin/ksh
[622]	1076	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1077	#PBS -l ${qsubmem}=${Memory}gb
	1078	#PBS -o $remote_dayfile
	1079	#PBS -N palm
	1080	#PBS -j o
	1081
	1082	%%END%%
	1083
	1084	else
	1085	cat > $job_to_send << %%END%%
	1086	#!/bin/ksh
	1087	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1088	#PBS -o $remote_dayfile
	1089	#PBS -j o
	1090
	1091	%%END%%
	1092
	1093	fi
	1094
[251]	1095	elif [[ $remote_host = necriam ]]
[35]	1096	then
	1097
	1098	if (( nodes > 1 ))
	1099	then
	1100	cat > $job_to_send << %%END%%
	1101	#!/bin/ksh
	1102	#PBS -b $nodes
	1103	#PBS -o $remote_dayfile
[253]	1104	#PBS -N $job_name
[35]	1105	#PBS -j o
[251]	1106	#PBS -v MPIPROGINV=YES
[35]	1107
	1108	%%END%%
	1109
	1110	elif [[ $numprocs != 0 ]]
	1111	then
	1112	cat > $job_to_send << %%END%%
	1113	#!/bin/ksh
	1114	#PBS -q ${queue}
	1115	#PBS -o $remote_dayfile
[253]	1116	#PBS -N $job_name
[35]	1117	#PBS -j o
[251]	1118	#PBS -v MPIPROGINV=YES
[35]	1119
	1120	%%END%%
	1121
	1122	else
	1123	cat > $job_to_send << %%END%%
	1124	#!/bin/ksh
	1125	#PBS -o $remote_dayfile
	1126	#PBS -j o
	1127
	1128	%%END%%
	1129
	1130	fi
	1131
[1]	1132	elif [[ $remote_host = lctit ]]
	1133	then
	1134	cat > $job_to_send << %%END%%
[635]	1135	#!/bin/ksh
[1]	1136
[678]	1137	$module_calls
	1138
[1]	1139	%%END%%
	1140
	1141	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
[678]	1142	if [[ $tasks_per_node != $processes_per_node ]]
	1143	then
	1144	submcom="$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 $queue "
	1145	else
	1146	submcom="$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 "
	1147	fi
[1]	1148
	1149	else
	1150
	1151	cat > $job_to_send << %%END%%
	1152	# @\$-q ${queue}
	1153	# @\$-l${qsubtime} $timestring
	1154	# @\$-l${qsubmem} ${memory}mb
	1155	# @\$-o $remote_dayfile
	1156	# @\$-eo
	1157
	1158	%%END%%
	1159
	1160	fi
	1161
	1162
	1163	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	1164	# BEI EXIT VERANLASSEN
	1165	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	1166	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	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
[635]	1172	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec ]]
[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
	1186	elif [[ $remote_host = ibmy ]]
	1187	then
	1188	return_queue=serial
[325]	1189	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
	1190	then
[326]	1191	return_queue=serialq
[251]	1192	elif [[ $remote_host = necriam ]]
[1]	1193	then
[253]	1194	return_queue=SP
[1]	1195	else
	1196	return_queue=unknown
	1197	fi
	1198
	1199	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1200	then
	1201
[622]	1202	if [[ $remote_host = ibmku ]]
	1203	then
	1204	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1205	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1206	else
	1207	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1208	fi
[1]	1209	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	1210	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	1211	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	1212	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	1213	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	1214	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
[312]	1215	if [[ $host != "ibmh" ]]
	1216	then
	1217	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	1218	fi
[1]	1219	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	1220	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	1221
	1222	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	1223	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	1224
	1225	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	1226	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]	1227	if [[ $remote_host = ibmku ]]
	1228	then
	1229	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
	1230	fi
[1]	1231	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	1232
	1233	elif [[ $remote_host = nech ]]
	1234	then
	1235	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1236	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1237	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1238	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1239	echo "#PBS -j o" >> $job_to_send
	1240	echo " " >> $job_to_send
	1241	echo "set -x" >> $job_to_send
	1242	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1243	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1244	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1245	echo "%%END%%" >> $job_to_send
	1246
[251]	1247	elif [[ $remote_host = necriam ]]
[35]	1248	then
	1249	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1250	echo "#PBS -q $return_queue" >> $job_to_send
	1251	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1252	echo "#PBS -j o" >> $job_to_send
	1253	echo " " >> $job_to_send
	1254	echo "set -x" >> $job_to_send
	1255	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1256	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1257	echo "%%END%%" >> $job_to_send
	1258
[181]	1259	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1260	then
	1261	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1262	echo "#!/bin/bash" >> $job_to_send
	1263	echo "#PBS -N job_protocol_transfer" >> $job_to_send
[325]	1264	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[181]	1265	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[366]	1266	echo "#PBS -l feature=data" >> $job_to_send
[181]	1267	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1268	echo "#PBS -j oe" >> $job_to_send
	1269	echo " " >> $job_to_send
	1270	echo ". /usr/share/modules/init/bash" >> $job_to_send
	1271	echo "set -x" >> $job_to_send
	1272	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1273	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1274	echo "%%END%%" >> $job_to_send
	1275
[1]	1276	else
	1277
	1278	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1279	echo "# @\\\$-q $return_queue" >> $job_to_send
	1280	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1281	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1282	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1283	then
	1284	echo "# @\$-l mpp_p=0" >> $job_to_send
	1285	fi
	1286	echo '# @\$-lF 10mb' >> $job_to_send
	1287	# echo '# @\$-o /dev/null' >> $job_to_send
	1288	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1289	echo '# @\\\$-eo' >> $job_to_send
	1290	echo " " >> $job_to_send
	1291	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1292	then
	1293	echo "set +vx" >> $job_to_send
	1294	echo ". .profile" >> $job_to_send
	1295	fi
	1296	echo "set -x" >> $job_to_send
	1297	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	1298	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1299	echo "%%END%%" >> $job_to_send
	1300	fi
	1301
	1302	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1303	then
	1304	echo "llsubmit scpjob.$kennung" >> $job_to_send
[181]	1305	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1306	then
[748]	1307	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1308	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1309	echo "msub scpjob.$kennung" >> $job_to_send
[1]	1310	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1311	then
	1312	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	1313	elif [[ $remote_host = t3es ]]
	1314	then
	1315	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
	1316	else
	1317	echo "qsub scpjob.$kennung" >> $job_to_send
	1318	fi
[622]	1319	if [[ $remote_host != ibmku ]]
	1320	then
	1321	echo "rm scpjob.$kennung" >> $job_to_send
	1322	fi
[1]	1323	if [[ $remote_host = nech ]]
	1324	then
	1325	echo "cd -" >> $job_to_send
	1326	fi
	1327	else
	1328	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1329	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1330	fi
	1331	echo "set -x" >> $job_to_send
	1332	echo " ' exit" >> $job_to_send
	1333	echo "set -x" >> $job_to_send
	1334	fi
	1335
	1336
	1337
	1338	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	1339	cat $file_to_send >> $job_to_send
[69]	1340
[1]	1341	if [[ $remote_host = ibm ]]
	1342	then
	1343	echo " " >> $job_to_send
	1344	echo "exit" >> $job_to_send
	1345	fi
[635]	1346
	1347	# remove job file
[622]	1348	if [[ $remote_host = lctit \|\| $remote_host = ibmku ]]
[69]	1349	then
	1350	echo " " >> $job_to_send
	1351	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1352	fi
[1]	1353
	1354
	1355
	1356	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	1357	if [[ -z $remote_user ]]
	1358	then
	1359	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	1360	then
	1361	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	1362	fi
	1363	fi
	1364
	1365
	1366
	1367	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	1368	if [[ $no_submit = false ]]
	1369	then
	1370	if [[ $remote_host != $local_host ]]
	1371	then
	1372	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	1373	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	1374	then
[82]	1375	job_catalog_save=$job_catalog
	1376	job_catalog=job_queue
	1377	elif [[ $remote_host = nech ]]
	1378	then
	1379	job_catalog_save=$job_catalog
	1380	job_catalog=/hpf/b/${remote_user}/job_queue
	1381	fi
	1382	if [[ $local_host = decalpha ]]
	1383	then
	1384	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	1385	# PASSWORT AB
	1386	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	1387	elif [[ $remote_host = nech ]]
	1388	then
	1389	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	1390	# TRANSFERIERT WERDEN
	1391	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1392	else
[82]	1393	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1394	fi
[82]	1395	if [[ $? = 1 ]]
	1396	then
	1397	locat=scp; exit
	1398	fi
	1399	if [[ $remote_host = ibms ]]
	1400	then
	1401	job_catalog=$job_catalog_save
	1402	fi
[1]	1403	[[ $verify = true ]] && printf "\n >>> finished\n"
	1404	else
	1405	eval job_catalog=$job_catalog
	1406	cp $job_to_send ${job_catalog}/$job_on_remhost
	1407	fi
	1408
	1409
	1410
	1411	# NQS- BZW. LOADLEVELER-JOB STARTEN
	1412	if [[ $remote_host != $local_host ]]
	1413	then
	1414	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1415
	1416	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
[1]	1417	then
[635]	1418	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
	1419	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
	1420	elif [[ $remote_host = ibmku ]]
	1421	then
	1422	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1]	1423	else
[635]	1424	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1425	fi
[1]	1426
	1427	[[ $verify = true ]] && printf " >>> o.k.\n"
	1428	else
	1429	cd $job_catalog
[210]	1430	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
[1]	1431	then
	1432	eval $submcom $job_on_remhost
[717]	1433	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 \|\| $localhost = lck \|\| $localhost = lckordi\|\| $localhost = lcyon ]]
[108]	1434	then
[635]	1435	chmod u+x $job_on_remhost
	1436	echo "$submcom $job_on_remhost"
[108]	1437	eval $submcom $job_on_remhost
[1]	1438	elif [[ $local_host = nech ]]
	1439	then
	1440	if [[ $queue = default ]]
	1441	then
	1442	eval $submcom $job_on_remhost
	1443	else
	1444	eval $submcom -q $queue $job_on_remhost
	1445	fi
	1446	else
	1447	qsub $job_on_remhost
	1448	fi
[622]	1449
	1450	# Jobfile must not be deleted on lctit/ibmku!! This will be done
	1451	# only at the end of the job.
	1452	if [[ $local_host != lctit && $local_host != ibmku ]]
	1453	then
	1454	rm $job_on_remhost
	1455	fi
[1]	1456	cd - > /dev/null
	1457	fi
	1458	fi
	1459
	1460
	1461
	1462	# ABSCHLUSSARBEITEN
	1463
	1464	if [[ $no_submit = false ]]
	1465	then
	1466	rm $job_to_send
	1467	fi
	1468	[[ $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 |