Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 313

Last change on this file since 313 was 312, checked in by raasch, 16 years ago
adjustments for IBM at DKRZ (ibmh)
Property svn:keywords set to `Id Rev`
File size: 41.9 KB

Rev	Line
[1]	1	#!/bin/ksh
[82]	2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
[169]	3	# $Id: subjob 312 2009-05-12 10:54:28Z 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
[1]	100
	101
[206]	102
[1]	103	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
	104	delete_dayfile=false
[122]	105	group_number=none
[1]	106	locat=normal
	107	no_default_queue=none
	108	no_submit=false
	109	job_catalog="~/job_queue"
	110	job_name=none
	111	local_user=$LOGNAME
	112	node_usage=shared
	113	numprocs=1
	114	punkte="..........................................................."
	115	submcom=qsub
	116	queue=default
	117	remote_host=none
	118	remote_user=""
	119	verify=true
	120
	121	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
	122	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
	123	typeset -L20 spalte1
	124	typeset -R40 spalte2
	125	typeset -L60 spalte3
	126
	127
	128
	129	# FEHLERBEHANDLUNG
	130	# BEI EXIT:
	131	trap 'if [[ $locat != normal ]]
	132	then
	133	case $locat in
	134	(option) printf "\n --> available optios can be displayed"
	135	printf " by typing:"
	136	printf "\n \"subjob ?\" \n";;
	137	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	138	(*) printf "\n +++ unknown error"
	139	printf "\n please inform S. Raasch!\n"
	140	esac
	141	[[ -f $job_to_send ]] && rm $job_to_send
	142	printf "\n\n+++ SUBJOB killed \n\n"
	143	fi' exit
	144
	145
	146	# BEI TERMINAL-BREAK:
	147	trap '[[ -f $job_to_send ]] && rm $job_to_send
	148	printf "\n\n+++ SUBJOB killed \n\n"
	149	exit
	150	' 2
	151
	152
	153
	154
	155	# LOKALEN HOSTNAMEN FESTSTELLEN
	156	local_host=$(hostname)
	157
	158
	159
	160	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
	161	# UEBERHAUPT ZULAESSIG IST
	162	case $local_host in
	163	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[102]	164	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	165	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[210]	166	(bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
[291]	167	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[1]	168	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
[12]	169	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[292]	170	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
	171	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[108]	172	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
[1]	173	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	174	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[108]	175	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
[101]	176	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[1]	177	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	178	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	179	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	180	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[164]	181	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
[292]	182	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
[210]	183	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
[312]	184	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[5]	185	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
	186	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
	187	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[164]	188	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
[1]	189	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	190	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	191	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	192	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[5]	193	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[1]	194	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
	195	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
[259]	196	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
[181]	197	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
[196]	198	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
[1]	199	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
	200	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
	201	(*) printf "\n +++ \"$local_host\" unknown";
	202	printf "\n please inform S. Raasch!";
	203	locat=parameter; exit;;
	204	esac
	205
	206
	207
	208	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	209	remote_host=$local_host
	210
	211
	212
	213
	214	# PROZEDUROPTIONEN EINLESEN
[122]	215	while getopts :c:dDg:h:m:n:N:O:q:t:T:u:vX: option
[1]	216	do
	217	case $option in
	218	(c) job_catalog=$OPTARG;;
	219	(d) delete_dayfile=true;;
	220	(D) no_submit=true;;
[125]	221	(g) group_number=$OPTARG;;
[1]	222	(h) remote_host=$OPTARG;;
	223	(m) memory=$OPTARG;;
	224	(n) job_name=$OPTARG;;
	225	(N) node_usage=$OPTARG;;
	226	(O) threads_per_task=$OPTARG;;
	227	(q) no_default_queue=$OPTARG;;
	228	(t) cputime=$OPTARG;;
	229	(T) tasks_per_node=$OPTARG;;
	230	(u) remote_user=$OPTARG;;
	231	(v) verify=false;;
	232	(X) numprocs=$OPTARG;;
	233	(\?) printf "\n +++ Option $OPTARG unknown \n";
	234	locat=option; exit;;
	235	esac
	236	done
	237
	238
	239	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	240	shift OPTIND-1; file_to_send=$1
	241
	242
	243	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	244	if [ "$1" = "?" ]
	245	then
	246	(printf "\n *** subjob can be called as follows:\n"
	247	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	248	printf "\n Description of available options:\n"
	249	printf "\n Option Description Default-Value"
	250	printf "\n -c job-input- and output-catalog ~/job_queue"
	251	printf "\n -d no job-protocol will be created ---"
	252	printf "\n -D only the job-file will be created ---"
	253	printf "\n -h execution host, available hosts: $remote_host"
[5]	254	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
[251]	255	printf "\n lctit, nech, necriam, unics"
[1]	256	printf "\n -m memory demand per process in MByte ---"
	257	printf "\n -n jobname <jobdatei>"
	258	printf "\n -O threads per task (for OpenMP usage) 1"
	259	printf "\n -q job-queue to be used default"
	260	printf "\n -t allowed cpu-time in seconds ---"
	261	printf "\n -T tasks per node (on parallel hosts) ---"
	262	printf "\n -u username on execution host from .netrc"
	263	printf "\n -v no prompt for confirmation ---"
	264	printf "\n -X # of processors (on parallel hosts) 1"
	265	printf "\n "
	266	printf "\n The only possible positional parameter is <jobfile>:"
	267	printf "\n The complete NQS-job must be provided here."
	268	printf "\n <jobfile>=? creates this outline\n\n") \| more
	269	exit
	270	fi
	271
	272
	273
	274	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	275	if [[ "$file_to_send" = "" ]]
	276	then
	277	printf "\n +++ job-file missing"
	278	locat=parameter; exit
	279	else
	280	if [[ -f $file_to_send ]]
	281	then
	282	true
	283	else
	284	printf "\n +++ job-file: "
	285	printf "\n $file_to_send"
	286	printf "\n does not exist"
	287	locat=parameter; exit
	288	fi
	289	fi
	290
	291
	292
	293	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	294	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	295	if [[ $job_name = none ]]
	296	then
	297	job_name=$file_to_send
	298	fi
	299	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	300	then
	301	printf "\n +++ job-file name: "
	302	printf "\n $job_name"
	303	printf "\n must not contain \"/\"-characters"
	304	locat=parameter; exit
	305	fi
	306
	307
	308
	309
	310	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	311	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	312	if [[ $remote_host = none ]]
	313	then
	314	printf "\n +++ host missing"
	315	locat=option; exit
	316	else
	317	case $remote_host in
[251]	318	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	319	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[312]	320	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[251]	321	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	322	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	323	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
[291]	324	(lcsgib) queue=smallq; remote_addres=130.73.232.103; submcom=/opt/moab/bin/msub;;
[251]	325	(lcsgih) queue=smallq; remote_addres=130.75.4.103; submcom=/opt/moab/bin/msub;;
	326	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
	327	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
	328	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	329	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
	330	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	331	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	332	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	333	locat=parameter; exit;;
[1]	334	esac
	335	fi
	336
	337
	338	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	339	if [[ $no_default_queue != none ]]
	340	then
	341	error=false
	342	ndq=$no_default_queue
	343	case $remote_host in
	344	(ibm) case $ndq in
	345	(p690_express\|p690_standard\|p690_long) error=false;;
	346	(*) error=true;;
	347	esac;;
	348	(ibmb) case $ndq in
	349	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
	350	(*) error=true;;
	351	esac;;
	352	(ibmh) case $ndq in
[312]	353	(no_class) error=false;;
[1]	354	(*) error=true;;
	355	esac;;
	356	(ibms) case $ndq in
	357	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	358	(*) error=true;;
	359	esac;;
	360	(ibmy) case $ndq in
	361	(parallel) error=false;;
	362	(*) error=true;;
	363	esac;;
[191]	364	(lcsgib\|lcsgih) case $ndq in
	365	(testq\|smallq\|bigq) error=false;;
	366	(*) error=true;;
	367	esac;;
[1]	368	(lctit) case $ndq in
[129]	369	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
[1]	370	(*) error=true;;
	371	esac;;
	372	(t3eb) case $ndq in
	373	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	374	(*) error=true;;
	375	esac;;
[251]	376	(necriam) case $ndq in
[253]	377	(SP\|SS\|P6) error=false;;
[251]	378	(*) error=true;;
	379	esac;;
[1]	380	(t3eh) case $ndq in
	381	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	382	(*) error=true;;
	383	esac;;
	384	(t3ej2\|t3ej5) case $ndq in
	385	(low\|normal\|high) error=false;;
	386	(*) error=true;;
	387	esac;;
	388	(t3es) case $ndq in
	389	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	390	(*) error=true;;
	391	esac;;
	392	(unics) case $ndq in
	393	(unics\|ht) error=false;;
	394	(*) error=true;;
	395	esac;;
	396	esac
	397	if [[ $error = true ]]
	398	then
	399	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	400	locat=parameter; exit
	401	else
	402	queue=$no_default_queue
	403	fi
	404	fi
	405
	406
	407
	408	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
	409	if [[ $node_usage != shared && $queue = cdev ]]
	410	then
	411	node_usage=shared
	412	fi
	413
	414
	415
	416	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	417	# AUFTEILEN
	418	done=false
	419	while [[ $done = false ]]
	420	do
	421	if (( $cputime <= 0 ))
	422	then
	423	printf "\n +++ wrong cpu-time or cpu-time missing"
	424	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	425	printf "\n >>> "
	426	read cputime 1>/dev/null 2>&1
	427	else
	428	done=true
	429	fi
	430	done
	431	if [[ $remote_host = nech ]]
	432	then
	433	if (( tasks_per_node != 0 ))
	434	then
	435	(( cputime = cputime * tasks_per_node ))
	436	elif [[ $numprocs != 0 ]]
	437	then
	438	(( cputime = cputime * numprocs ))
	439	fi
	440	fi
	441	(( stunden = cputime / 3600 ))
	442	(( resttime = cputime - stunden * 3600 ))
	443	(( minuten = resttime / 60 ))
	444	(( sekunden = resttime - minuten * 60 ))
	445	timestring=${stunden}:${minuten}:${sekunden}
	446
	447
	448
	449	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	450	done=false
	451	while [[ $done = false ]]
	452	do
	453	if (( memory <= 0 ))
	454	then
	455	printf "\n +++ wrong memory demand or memory demand missing"
	456	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	457	printf "\n >>> "
	458	read memory 1>/dev/null 2>&1
	459	else
	460	done=true
	461	fi
	462	done
	463
[251]	464	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
[1]	465	then
	466	if (( tasks_per_node != 0 ))
	467	then
	468	(( Memory = memory * tasks_per_node / 1000 ))
	469	elif [[ $numprocs != 0 ]]
	470	then
	471	(( Memory = memory * numprocs / 1000 ))
	472	else
	473	(( Memory = memory / 1000 ))
	474	fi
	475	fi
	476
	477
	478	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	479	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	480	then
	481	(( memory = memory * threads_per_task ))
	482	fi
	483
	484
	485	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	486	if (( tasks_per_node != 0 ))
	487	then
	488	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
	489	fi
	490
	491
	492
	493	# HEADER-AUSGABE
	494	if [[ $verify = true ]]
	495	then
	496	printf "\n\n"
	497	printf "#--------------------------------------------------------------# \n"
	498	spalte1=SUBJOB;spalte2=$(date)
	499	printf "\| $spalte1$spalte2 \| \n"
	500	printf "\| \| \n"
	501	printf "\| values of parameters/options: \| \n"
	502	spalte1=local_host$punkte; spalte2=$punkte$local_host
	503	printf "\| $spalte1$spalte2 \| \n"
	504	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	505	printf "\| $spalte1$spalte2 \| \n"
	506	spalte1=queue$punkte; spalte2=$punkte$queue
	507	printf "\| $spalte1$spalte2 \| \n"
	508	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	509	printf "\| $spalte1$spalte2 \| \n"
	510	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	511	printf "\| $spalte1$spalte2 \| \n"
	512	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	513	printf "\| $spalte1$spalte2 \| \n"
	514	printf "#--------------------------------------------------------------# \n\n"
	515
	516
	517	# KONTROLLABFRAGE, OB ALLES O.K.
	518	antwort="dummy"
	519	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	520	do
	521	read antwort?" >>> continue (y/n) ? "
	522	done
	523	if [[ $antwort = n \|\| $antwort = N ]]
	524	then
	525	locat=verify; exit
	526	fi
	527	printf "\n"
	528	fi
	529
	530
	531
	532	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	533	kennung=$RANDOM
	534	job_on_remhost=${job_name}_${kennung}_$local_host
	535	job_to_send=job_to_send_$kennung
	536	if [[ $delete_dayfile = false ]]
	537	then
	538	remote_dayfile=${local_host}_${job_name}_result_$kennung
	539	local_dayfile=${remote_host}_${job_name}
	540	else
	541	remote_dayfile=/dev/null
	542	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	543	then
	544	notification='# @ notification = never'
	545	fi
	546	fi
	547
	548
	549
	550	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
	551	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	552	then
	553
	554	if [[ $remote_host = ibmy ]]
	555	then
	556	consumable_memory=""
	557	else
	558	consumable_memory="ConsumableMemory($memory mb)"
	559	fi
[312]	560
	561	if [[ $remote_host = ibmh ]]
[1]	562	then
[312]	563	data_limit=""
	564	network_to_use=""
	565	class=""
	566	environment=""
[1]	567	else
[312]	568	class="# @ class = $queue"
	569	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	570	if [[ $queue = cdev ]]
[1]	571	then
[312]	572	data_limit="# @ data_limit = 1.76gb"
	573	network_to_use="# @ network.mpi = sn_all,shared,ip"
[1]	574	else
[312]	575	if [[ $remote_host = ibms ]]
	576	then
	577	network_to_use="# @ network.mpi = csss,shared,us"
	578	elif [[ $remote_host = ibmy ]]
	579	then
	580	network_to_use=""
	581	else
	582	network_to_use="# @ network.mpi = sn_all,shared,us"
	583	data_limit="# @ data_limit = 1.76gb"
	584	fi
[1]	585	fi
	586	fi
	587
	588	cat > $job_to_send << %%END%%
	589	#!/bin/ksh
[312]	590	# @ shell = /bin/ksh
[1]	591
	592	# @ job_type = parallel
	593	# @ job_name = palm
	594	# @ wall_clock_limit = ${timestring},$timestring
	595	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	596	# @ output = $remote_dayfile
	597	# @ error = $remote_dayfile
	598	# @ image_size = 50
[312]	599	$class
	600	$environment
[1]	601	$network_to_use
	602	$data_limit
	603	$notification
	604
	605	%%END%%
	606
	607	if (( nodes > 0 ))
	608	then
	609
	610	cat >> $job_to_send << %%END%%
	611	# @ node = $nodes
	612	# @ tasks_per_node = $tasks_per_node
	613	# @ node_usage = $node_usage
	614	# @ queue
	615
	616	%%END%%
	617
	618	else
	619
	620	if [[ $remote_host != ibmy ]]
	621	then
	622
	623	cat >> $job_to_send << %%END%%
	624	# @ blocking = unlimited
	625	# @ total_tasks = $numprocs
	626	# @ node_usage = $node_usage
	627	# @ queue
	628
	629	%%END%%
	630
	631	else
	632
	633	cat >> $job_to_send << %%END%%
	634	# @ node = 1
	635	# @ total_tasks = $numprocs
	636	# @ queue
	637
	638	%%END%%
	639
	640	fi
	641
	642	fi
	643
	644	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	645	then
	646
	647	cat > $job_to_send << %%END%%
	648	#!/bin/ksh
	649
	650	# @ job_type = serial
	651	# @ node_usage = $node_usage
	652	# @ job_name = palm
	653	# @ wall_clock_limit = ${timestring},$timestring
	654	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	655	# @ output = $remote_dayfile
	656	# @ error = $remote_dayfile
[312]	657	$class
[1]	658	$notification
	659
	660	# @ queue
	661
	662	%%END%%
	663
[108]	664	elif [[ $remote_host = lcfimm ]]
	665	then
	666
	667	if [[ $numprocs != 0 ]]
	668	then
	669	cat > $job_to_send << %%END%%
	670	#!/bin/ksh
	671	#PBS -N $job_name
[206]	672	#PBS -A geofysisk
[108]	673	#PBS -l walltime=$timestring
	674	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
	675	#PBS -l pmem=${memory}mb
	676	#PBS -m abe
[206]	677	#PBS -M bjorn.maronga@student.uib.no
[108]	678	#PBS -o $remote_dayfile
	679	#PBS -j oe
	680	mpd &
	681
	682	%%END%%
	683
[168]	684	else
	685	cat > $job_to_send << %%END%%
	686	#!/bin/ksh
	687	#PBS -N $job_name
[206]	688	#PBS -A geofysisk
[168]	689	#PBS -l walltime=$timestring
	690	#PBS -l ncpus=1
	691	#PBS -l pmem=${memory}mb
	692	#PBS -m abe
[206]	693	#PBS -M bjorn.maronga@student.uib.no
[168]	694	#PBS -o $remote_dayfile
	695	#PBS -j oe
	696
	697	%%END%%
	698
	699	fi
	700
[181]	701	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
[179]	702	then
	703
[191]	704	if [[ $queue = testq ]]
	705	then
	706	queue_directive="#PBS -q $queue"
	707	else
	708	queue_directive=""
	709	fi
	710
[179]	711	if [[ $numprocs != 0 ]]
	712	then
	713	cat > $job_to_send << %%END%%
[181]	714	#!/bin/bash
[179]	715	#PBS -N $job_name
	716	#PBS -l walltime=$timestring
[181]	717	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
	718	#PBS -l feature=$sgi_feature
	719	#PBS -l naccesspolicy=$node_usage
[179]	720	#PBS -o $remote_dayfile
	721	#PBS -j oe
[191]	722	$queue_directive
[179]	723
	724	. /usr/share/modules/init/bash
[221]	725	module load $mpilib
[196]	726	# module load mvapich2/1.2rc1-intel
	727	# module load mvapich2
[179]	728	module load netcdf
[181]	729	export LD_LIBRARY_PATH=/sw/intel/cce/10.1.015/lib:/sw/intel/fce/10.1.015/lib:/sw/dataformats/netcdf/3.6.2/lib:\$LD_LIBRARY_PATH
	730	echo ld_library_path=\$LD_LIBRARY_PATH
[179]	731
	732	%%END%%
	733
	734	else
	735	cat > $job_to_send << %%END%%
	736	#PBS -S /bin/bash
	737	#PBS -N $job_name
	738	#PBS -l walltime=$timestring
	739	#PBS -l ncpus=1
	740	#PBS -o $remote_dayfile
	741	#PBS -e $remote_dayfile
	742
	743	%%END%%
	744
	745	fi
	746
[164]	747	elif [[ $remote_host = lcxt4 ]]
	748	then
	749
	750	if [[ $numprocs != 0 ]]
	751	then
	752	cat > $job_to_send << %%END%%
	753	#!/bin/ksh
	754	#PBS -S /bin/ksh
	755	#PBS -N $job_name
[206]	756	#PBS -A geofysisk
	757	#PBS -j oe
[164]	758	#PBS -l walltime=$timestring
	759	#PBS -l mppwidth=${numprocs}
	760	#PBS -l mppnppn=${tasks_per_node}
	761	#PBS -m abe
[206]	762	#PBS -M bjorn.maronga@student.uib.no
[164]	763	#PBS -o $remote_dayfile
	764
	765	%%END%%
	766
[108]	767	else
	768	cat > $job_to_send << %%END%%
	769	#!/bin/ksh
[168]	770	#PBS -S /bin/ksh
[108]	771	#PBS -N $job_name
[206]	772	#PBS -A geofysisk
	773	#PBS -j oe
[108]	774	#PBS -l walltime=$timestring
	775	#PBS -l ncpus=1
	776	#PBS -l pmem=${memory}mb
	777	#PBS -m abe
[206]	778	#PBS -M bjorn.maronga@student.uib.no
[108]	779	#PBS -o $remote_dayfile
	780
	781	%%END%%
	782
	783	fi
	784
[1]	785	elif [[ $remote_host = nech ]]
	786	then
	787
	788	if (( nodes > 1 ))
	789	then
	790	cat > $job_to_send << %%END%%
	791	#!/bin/ksh
	792	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	793	#PBS -l ${qsubmem}=${Memory}gb
	794	#PBS -b $nodes
	795	#PBS -o $remote_dayfile
	796	#PBS -N palm
	797	#PBS -j o
	798	#PBS -T mpisx
	799
	800	%%END%%
	801
	802	elif [[ $numprocs != 0 ]]
	803	then
	804	cat > $job_to_send << %%END%%
	805	#!/bin/ksh
	806	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	807	#PBS -l ${qsubmem}=${Memory}gb
	808	#PBS -o $remote_dayfile
	809	#PBS -N palm
	810	#PBS -j o
	811
	812	%%END%%
	813
	814	else
	815	cat > $job_to_send << %%END%%
	816	#!/bin/ksh
	817	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	818	#PBS -o $remote_dayfile
	819	#PBS -j o
	820
	821	%%END%%
	822
	823	fi
	824
[251]	825	elif [[ $remote_host = necriam ]]
[35]	826	then
	827
	828	if (( nodes > 1 ))
	829	then
	830	cat > $job_to_send << %%END%%
	831	#!/bin/ksh
	832	#PBS -b $nodes
	833	#PBS -o $remote_dayfile
[253]	834	#PBS -N $job_name
[35]	835	#PBS -j o
[251]	836	#PBS -v MPIPROGINV=YES
[35]	837
	838	%%END%%
	839
	840	elif [[ $numprocs != 0 ]]
	841	then
	842	cat > $job_to_send << %%END%%
	843	#!/bin/ksh
	844	#PBS -q ${queue}
	845	#PBS -o $remote_dayfile
[253]	846	#PBS -N $job_name
[35]	847	#PBS -j o
[251]	848	#PBS -v MPIPROGINV=YES
[35]	849
	850	%%END%%
	851
	852	else
	853	cat > $job_to_send << %%END%%
	854	#!/bin/ksh
	855	#PBS -o $remote_dayfile
	856	#PBS -j o
	857
	858	%%END%%
	859
	860	fi
	861
[1]	862	elif [[ $remote_host = lctit ]]
	863	then
	864	cat > $job_to_send << %%END%%
	865	#!/bin/bash
	866	#$ -S /bin/bash
	867	cd $job_catalog
[69]	868	export PATH=$PALM_BIN:\$PATH
	869	echo \$PATH
[1]	870
	871	%%END%%
	872
	873	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
[118]	874	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
[122]	875	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
[1]	876
	877	else
	878
	879	cat > $job_to_send << %%END%%
	880	# @\$-q ${queue}
	881	# @\$-l${qsubtime} $timestring
	882	# @\$-l${qsubmem} ${memory}mb
	883	# @\$-o $remote_dayfile
	884	# @\$-eo
	885
	886	%%END%%
	887
	888	fi
	889
	890
	891	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	892	# BEI EXIT VERANLASSEN
	893	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	894	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	895	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	896	then
	897	echo "set +vx" >> $job_to_send
	898	echo "trap '" >> $job_to_send
	899	echo "set +vx" >> $job_to_send
[181]	900	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
[1]	901	then
	902	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
	903	then
	904	return_queue=c1
	905	elif [[ $remote_host = ibms ]]
	906	then
	907	return_queue=p_normal
	908	elif [[ $remote_host = ibmy ]]
	909	then
	910	return_queue=serial
	911	elif [[ $remote_host = lctit ]]
	912	then
[118]	913	return_queue=sla3
[251]	914	elif [[ $remote_host = necriam ]]
[1]	915	then
[253]	916	return_queue=SP
[1]	917	else
	918	return_queue=unknown
	919	fi
	920
	921	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	922	then
	923
	924	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	925	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	926	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	927	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	928	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	929	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	930	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
[312]	931	if [[ $host != "ibmh" ]]
	932	then
	933	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	934	fi
[1]	935	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	936	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	937
	938	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	939	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	940
	941	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	942	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
	943	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	944
	945	elif [[ $remote_host = nech ]]
	946	then
	947	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	948	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	949	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	950	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	951	echo "#PBS -j o" >> $job_to_send
	952	echo " " >> $job_to_send
	953	echo "set -x" >> $job_to_send
	954	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	955	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	956	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	957	echo "%%END%%" >> $job_to_send
	958
[251]	959	elif [[ $remote_host = necriam ]]
[35]	960	then
	961	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	962	echo "#PBS -q $return_queue" >> $job_to_send
	963	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	964	echo "#PBS -j o" >> $job_to_send
	965	echo " " >> $job_to_send
	966	echo "set -x" >> $job_to_send
	967	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	968	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	969	echo "%%END%%" >> $job_to_send
	970
[181]	971	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	972	then
	973	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	974	echo "#!/bin/bash" >> $job_to_send
	975	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	976	echo "#PBS -l walltime=00:02:00" >> $job_to_send
	977	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
	978	echo "#PBS -l feature=xe" >> $job_to_send
	979	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	980	echo "#PBS -j oe" >> $job_to_send
	981	echo " " >> $job_to_send
	982	echo ". /usr/share/modules/init/bash" >> $job_to_send
	983	echo "set -x" >> $job_to_send
	984	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	985	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	986	echo "%%END%%" >> $job_to_send
	987
[1]	988	elif [[ $remote_host = lctit ]]
	989	then
	990	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	991	echo "set -x" >> $job_to_send
	992	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
	993	echo "cd $job_catalog" >> $job_to_send
	994	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	995	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	996	echo "%%END%%" >> $job_to_send
	997
	998	else
	999
	1000	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1001	echo "# @\\\$-q $return_queue" >> $job_to_send
	1002	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1003	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1004	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1005	then
	1006	echo "# @\$-l mpp_p=0" >> $job_to_send
	1007	fi
	1008	echo '# @\$-lF 10mb' >> $job_to_send
	1009	# echo '# @\$-o /dev/null' >> $job_to_send
	1010	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1011	echo '# @\\\$-eo' >> $job_to_send
	1012	echo " " >> $job_to_send
	1013	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1014	then
	1015	echo "set +vx" >> $job_to_send
	1016	echo ". .profile" >> $job_to_send
	1017	fi
	1018	echo "set -x" >> $job_to_send
	1019	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	1020	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1021	echo "%%END%%" >> $job_to_send
	1022	fi
	1023
	1024	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1025	then
	1026	echo "llsubmit scpjob.$kennung" >> $job_to_send
[181]	1027	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1028	then
	1029	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1030	echo "msub scpjob.$kennung" >> $job_to_send
[1]	1031	elif [[ $remote_host = lctit ]]
	1032	then
	1033	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1034	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
	1035	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
	1036	echo "sleep 10" >> $job_to_send
	1037	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1038	then
	1039	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	1040	elif [[ $remote_host = t3es ]]
	1041	then
	1042	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
	1043	else
	1044	echo "qsub scpjob.$kennung" >> $job_to_send
	1045	fi
	1046	echo "rm scpjob.$kennung" >> $job_to_send
	1047	if [[ $remote_host = nech ]]
	1048	then
	1049	echo "cd -" >> $job_to_send
	1050	fi
	1051	else
	1052	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1053	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1054	fi
	1055	echo "set -x" >> $job_to_send
	1056	echo " ' exit" >> $job_to_send
	1057	echo "set -x" >> $job_to_send
	1058	fi
	1059
	1060
	1061
	1062	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	1063	cat $file_to_send >> $job_to_send
[69]	1064
[1]	1065	if [[ $remote_host = ibm ]]
	1066	then
	1067	echo " " >> $job_to_send
	1068	echo "exit" >> $job_to_send
	1069	fi
[69]	1070	if [[ $remote_host = lctit ]]
	1071	then
	1072	echo " " >> $job_to_send
	1073	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1074	fi
[1]	1075
	1076
	1077
	1078	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	1079	if [[ -z $remote_user ]]
	1080	then
	1081	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	1082	then
	1083	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	1084	fi
	1085	fi
	1086
	1087
	1088
	1089	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	1090	if [[ $no_submit = false ]]
	1091	then
	1092	if [[ $remote_host != $local_host ]]
	1093	then
	1094	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	1095	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	1096	then
[82]	1097	job_catalog_save=$job_catalog
	1098	job_catalog=job_queue
	1099	elif [[ $remote_host = nech ]]
	1100	then
	1101	job_catalog_save=$job_catalog
	1102	job_catalog=/hpf/b/${remote_user}/job_queue
	1103	fi
	1104	if [[ $local_host = decalpha ]]
	1105	then
	1106	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	1107	# PASSWORT AB
	1108	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	1109	elif [[ $remote_host = nech ]]
	1110	then
	1111	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	1112	# TRANSFERIERT WERDEN
	1113	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1114	else
[82]	1115	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1116	fi
[82]	1117	if [[ $? = 1 ]]
	1118	then
	1119	locat=scp; exit
	1120	fi
	1121	if [[ $remote_host = ibms ]]
	1122	then
	1123	job_catalog=$job_catalog_save
	1124	fi
[1]	1125	[[ $verify = true ]] && printf "\n >>> finished\n"
	1126	else
	1127	eval job_catalog=$job_catalog
	1128	cp $job_to_send ${job_catalog}/$job_on_remhost
	1129	fi
	1130
	1131
	1132
	1133	# NQS- BZW. LOADLEVELER-JOB STARTEN
	1134	if [[ $remote_host != $local_host ]]
	1135	then
	1136	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[82]	1137	if [[ $remote_host != lctit ]]
[1]	1138	then
[191]	1139	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[1]	1140	else
[206]	1141	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
[82]	1142	# MIT SSH, DESHALB AUFRUF PER PIPE
	1143	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
	1144	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	1145	echo "first try" > send_protocol
	1146	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
	1147	do
	1148	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
	1149	then
	1150	printf "\n trying to submit job to TIT cluster..."
	1151	else
	1152	printf "\n +++ failed ... trying to submit again ..."
	1153	fi
	1154	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
	1155	cat send_protocol
	1156	done
	1157	sleep 10
	1158	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	1159	echo "$submcom $job_on_remhost"
	1160	rm send_protocol
	1161	fi
[1]	1162
	1163	[[ $verify = true ]] && printf " >>> o.k.\n"
	1164	else
	1165	cd $job_catalog
[210]	1166	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
[1]	1167	then
	1168	eval $submcom $job_on_remhost
[164]	1169	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 ]]
[108]	1170	then
	1171	eval $submcom $job_on_remhost
	1172	echo "$submcom $job_on_remhost"
	1173	chmod u+x $job_on_remhost
[292]	1174	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
	1175	then
	1176	eval $submcom $job_on_remhost
[1]	1177	elif [[ $local_host = nech ]]
	1178	then
	1179	if [[ $queue = default ]]
	1180	then
	1181	eval $submcom $job_on_remhost
	1182	else
	1183	eval $submcom -q $queue $job_on_remhost
	1184	fi
	1185	else
	1186	qsub $job_on_remhost
	1187	fi
[69]	1188	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
	1189	[[ $local_host != lctit ]] && rm $job_on_remhost
[1]	1190	cd - > /dev/null
	1191	fi
	1192	fi
	1193
	1194
	1195
	1196	# ABSCHLUSSARBEITEN
	1197
	1198	if [[ $no_submit = false ]]
	1199	then
	1200	rm $job_to_send
	1201	fi
	1202	[[ $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 |