Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 335

Last change on this file since 335 was 326, checked in by raasch, 16 years ago
bugfix in subjob
Property svn:keywords set to `Id Rev`
File size: 42.1 KB

Rev	Line
[1]	1	#!/bin/ksh
[82]	2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
[169]	3	# $Id: subjob 326 2009-05-20 12:40:29Z heinze $
[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
[1]	101
	102
[206]	103
[1]	104	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
	105	delete_dayfile=false
[122]	106	group_number=none
[1]	107	locat=normal
	108	no_default_queue=none
	109	no_submit=false
	110	job_catalog="~/job_queue"
	111	job_name=none
	112	local_user=$LOGNAME
	113	node_usage=shared
	114	numprocs=1
	115	punkte="..........................................................."
	116	submcom=qsub
	117	queue=default
	118	remote_host=none
	119	remote_user=""
	120	verify=true
	121
	122	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
	123	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
	124	typeset -L20 spalte1
	125	typeset -R40 spalte2
	126	typeset -L60 spalte3
	127
	128
	129
	130	# FEHLERBEHANDLUNG
	131	# BEI EXIT:
	132	trap 'if [[ $locat != normal ]]
	133	then
	134	case $locat in
	135	(option) printf "\n --> available optios can be displayed"
	136	printf " by typing:"
	137	printf "\n \"subjob ?\" \n";;
	138	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	139	(*) printf "\n +++ unknown error"
	140	printf "\n please inform S. Raasch!\n"
	141	esac
	142	[[ -f $job_to_send ]] && rm $job_to_send
	143	printf "\n\n+++ SUBJOB killed \n\n"
	144	fi' exit
	145
	146
	147	# BEI TERMINAL-BREAK:
	148	trap '[[ -f $job_to_send ]] && rm $job_to_send
	149	printf "\n\n+++ SUBJOB killed \n\n"
	150	exit
	151	' 2
	152
	153
	154
	155
	156	# LOKALEN HOSTNAMEN FESTSTELLEN
	157	local_host=$(hostname)
	158
	159
	160
	161	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
	162	# UEBERHAUPT ZULAESSIG IST
	163	case $local_host in
	164	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[102]	165	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	166	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[210]	167	(bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
[291]	168	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[1]	169	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
[12]	170	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[292]	171	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
	172	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[108]	173	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
[1]	174	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	175	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[108]	176	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
[101]	177	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[1]	178	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	179	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	180	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	181	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[164]	182	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
[292]	183	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
[210]	184	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
[312]	185	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[5]	186	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
	187	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
	188	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[164]	189	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
[1]	190	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	191	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	192	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	193	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[5]	194	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[1]	195	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
	196	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
[259]	197	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
[181]	198	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
[196]	199	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
[1]	200	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
	201	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
	202	(*) printf "\n +++ \"$local_host\" unknown";
	203	printf "\n please inform S. Raasch!";
	204	locat=parameter; exit;;
	205	esac
	206
	207
	208
	209	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	210	remote_host=$local_host
	211
	212
	213
	214
	215	# PROZEDUROPTIONEN EINLESEN
[122]	216	while getopts :c:dDg:h:m:n:N:O:q:t:T:u:vX: option
[1]	217	do
	218	case $option in
	219	(c) job_catalog=$OPTARG;;
	220	(d) delete_dayfile=true;;
	221	(D) no_submit=true;;
[125]	222	(g) group_number=$OPTARG;;
[1]	223	(h) remote_host=$OPTARG;;
	224	(m) memory=$OPTARG;;
	225	(n) job_name=$OPTARG;;
	226	(N) node_usage=$OPTARG;;
	227	(O) threads_per_task=$OPTARG;;
	228	(q) no_default_queue=$OPTARG;;
	229	(t) cputime=$OPTARG;;
	230	(T) tasks_per_node=$OPTARG;;
	231	(u) remote_user=$OPTARG;;
	232	(v) verify=false;;
	233	(X) numprocs=$OPTARG;;
	234	(\?) printf "\n +++ Option $OPTARG unknown \n";
	235	locat=option; exit;;
	236	esac
	237	done
	238
	239
	240	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	241	shift OPTIND-1; file_to_send=$1
	242
	243
	244	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	245	if [ "$1" = "?" ]
	246	then
	247	(printf "\n *** subjob can be called as follows:\n"
	248	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	249	printf "\n Description of available options:\n"
	250	printf "\n Option Description Default-Value"
	251	printf "\n -c job-input- and output-catalog ~/job_queue"
	252	printf "\n -d no job-protocol will be created ---"
	253	printf "\n -D only the job-file will be created ---"
	254	printf "\n -h execution host, available hosts: $remote_host"
[5]	255	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
[251]	256	printf "\n lctit, nech, necriam, unics"
[1]	257	printf "\n -m memory demand per process in MByte ---"
	258	printf "\n -n jobname <jobdatei>"
	259	printf "\n -O threads per task (for OpenMP usage) 1"
	260	printf "\n -q job-queue to be used default"
	261	printf "\n -t allowed cpu-time in seconds ---"
	262	printf "\n -T tasks per node (on parallel hosts) ---"
	263	printf "\n -u username on execution host from .netrc"
	264	printf "\n -v no prompt for confirmation ---"
	265	printf "\n -X # of processors (on parallel hosts) 1"
	266	printf "\n "
	267	printf "\n The only possible positional parameter is <jobfile>:"
	268	printf "\n The complete NQS-job must be provided here."
	269	printf "\n <jobfile>=? creates this outline\n\n") \| more
	270	exit
	271	fi
	272
	273
	274
	275	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	276	if [[ "$file_to_send" = "" ]]
	277	then
	278	printf "\n +++ job-file missing"
	279	locat=parameter; exit
	280	else
	281	if [[ -f $file_to_send ]]
	282	then
	283	true
	284	else
	285	printf "\n +++ job-file: "
	286	printf "\n $file_to_send"
	287	printf "\n does not exist"
	288	locat=parameter; exit
	289	fi
	290	fi
	291
	292
	293
	294	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	295	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	296	if [[ $job_name = none ]]
	297	then
	298	job_name=$file_to_send
	299	fi
	300	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	301	then
	302	printf "\n +++ job-file name: "
	303	printf "\n $job_name"
	304	printf "\n must not contain \"/\"-characters"
	305	locat=parameter; exit
	306	fi
	307
	308
	309
	310
	311	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	312	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	313	if [[ $remote_host = none ]]
	314	then
	315	printf "\n +++ host missing"
	316	locat=option; exit
	317	else
	318	case $remote_host in
[251]	319	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	320	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[312]	321	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[251]	322	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	323	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	324	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
[291]	325	(lcsgib) queue=smallq; remote_addres=130.73.232.103; submcom=/opt/moab/bin/msub;;
[251]	326	(lcsgih) queue=smallq; remote_addres=130.75.4.103; submcom=/opt/moab/bin/msub;;
	327	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
	328	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
	329	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	330	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
	331	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	332	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	333	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	334	locat=parameter; exit;;
[1]	335	esac
	336	fi
	337
	338
	339	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	340	if [[ $no_default_queue != none ]]
	341	then
	342	error=false
	343	ndq=$no_default_queue
	344	case $remote_host in
	345	(ibm) case $ndq in
	346	(p690_express\|p690_standard\|p690_long) error=false;;
	347	(*) error=true;;
	348	esac;;
	349	(ibmb) case $ndq in
	350	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
	351	(*) error=true;;
	352	esac;;
	353	(ibmh) case $ndq in
[312]	354	(no_class) error=false;;
[1]	355	(*) error=true;;
	356	esac;;
	357	(ibms) case $ndq in
	358	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	359	(*) error=true;;
	360	esac;;
	361	(ibmy) case $ndq in
	362	(parallel) error=false;;
	363	(*) error=true;;
	364	esac;;
[191]	365	(lcsgib\|lcsgih) case $ndq in
[325]	366	(testq\|serialq\|smallq\|bigq) error=false;;
[191]	367	(*) error=true;;
	368	esac;;
[1]	369	(lctit) case $ndq in
[129]	370	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
[1]	371	(*) error=true;;
	372	esac;;
	373	(t3eb) case $ndq in
	374	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	375	(*) error=true;;
	376	esac;;
[251]	377	(necriam) case $ndq in
[253]	378	(SP\|SS\|P6) error=false;;
[251]	379	(*) error=true;;
	380	esac;;
[1]	381	(t3eh) case $ndq in
	382	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	383	(*) error=true;;
	384	esac;;
	385	(t3ej2\|t3ej5) case $ndq in
	386	(low\|normal\|high) error=false;;
	387	(*) error=true;;
	388	esac;;
	389	(t3es) case $ndq in
	390	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	391	(*) error=true;;
	392	esac;;
	393	(unics) case $ndq in
	394	(unics\|ht) error=false;;
	395	(*) error=true;;
	396	esac;;
	397	esac
	398	if [[ $error = true ]]
	399	then
	400	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	401	locat=parameter; exit
	402	else
	403	queue=$no_default_queue
	404	fi
	405	fi
	406
	407
	408
	409	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
	410	if [[ $node_usage != shared && $queue = cdev ]]
	411	then
	412	node_usage=shared
	413	fi
	414
	415
	416
	417	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	418	# AUFTEILEN
	419	done=false
	420	while [[ $done = false ]]
	421	do
	422	if (( $cputime <= 0 ))
	423	then
	424	printf "\n +++ wrong cpu-time or cpu-time missing"
	425	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	426	printf "\n >>> "
	427	read cputime 1>/dev/null 2>&1
	428	else
	429	done=true
	430	fi
	431	done
	432	if [[ $remote_host = nech ]]
	433	then
	434	if (( tasks_per_node != 0 ))
	435	then
	436	(( cputime = cputime * tasks_per_node ))
	437	elif [[ $numprocs != 0 ]]
	438	then
	439	(( cputime = cputime * numprocs ))
	440	fi
	441	fi
	442	(( stunden = cputime / 3600 ))
	443	(( resttime = cputime - stunden * 3600 ))
	444	(( minuten = resttime / 60 ))
	445	(( sekunden = resttime - minuten * 60 ))
	446	timestring=${stunden}:${minuten}:${sekunden}
	447
	448
	449
	450	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	451	done=false
	452	while [[ $done = false ]]
	453	do
	454	if (( memory <= 0 ))
	455	then
	456	printf "\n +++ wrong memory demand or memory demand missing"
	457	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	458	printf "\n >>> "
	459	read memory 1>/dev/null 2>&1
	460	else
	461	done=true
	462	fi
	463	done
	464
[251]	465	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
[1]	466	then
	467	if (( tasks_per_node != 0 ))
	468	then
	469	(( Memory = memory * tasks_per_node / 1000 ))
	470	elif [[ $numprocs != 0 ]]
	471	then
	472	(( Memory = memory * numprocs / 1000 ))
	473	else
	474	(( Memory = memory / 1000 ))
	475	fi
	476	fi
	477
	478
	479	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	480	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	481	then
	482	(( memory = memory * threads_per_task ))
	483	fi
	484
	485
	486	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	487	if (( tasks_per_node != 0 ))
	488	then
	489	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
	490	fi
	491
	492
	493
	494	# HEADER-AUSGABE
	495	if [[ $verify = true ]]
	496	then
	497	printf "\n\n"
	498	printf "#--------------------------------------------------------------# \n"
	499	spalte1=SUBJOB;spalte2=$(date)
	500	printf "\| $spalte1$spalte2 \| \n"
	501	printf "\| \| \n"
	502	printf "\| values of parameters/options: \| \n"
	503	spalte1=local_host$punkte; spalte2=$punkte$local_host
	504	printf "\| $spalte1$spalte2 \| \n"
	505	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	506	printf "\| $spalte1$spalte2 \| \n"
	507	spalte1=queue$punkte; spalte2=$punkte$queue
	508	printf "\| $spalte1$spalte2 \| \n"
	509	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	510	printf "\| $spalte1$spalte2 \| \n"
	511	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	512	printf "\| $spalte1$spalte2 \| \n"
	513	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	514	printf "\| $spalte1$spalte2 \| \n"
	515	printf "#--------------------------------------------------------------# \n\n"
	516
	517
	518	# KONTROLLABFRAGE, OB ALLES O.K.
	519	antwort="dummy"
	520	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	521	do
	522	read antwort?" >>> continue (y/n) ? "
	523	done
	524	if [[ $antwort = n \|\| $antwort = N ]]
	525	then
	526	locat=verify; exit
	527	fi
	528	printf "\n"
	529	fi
	530
	531
	532
	533	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	534	kennung=$RANDOM
	535	job_on_remhost=${job_name}_${kennung}_$local_host
	536	job_to_send=job_to_send_$kennung
	537	if [[ $delete_dayfile = false ]]
	538	then
	539	remote_dayfile=${local_host}_${job_name}_result_$kennung
	540	local_dayfile=${remote_host}_${job_name}
	541	else
	542	remote_dayfile=/dev/null
	543	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	544	then
	545	notification='# @ notification = never'
	546	fi
	547	fi
	548
	549
	550
	551	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
	552	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	553	then
	554
	555	if [[ $remote_host = ibmy ]]
	556	then
	557	consumable_memory=""
	558	else
	559	consumable_memory="ConsumableMemory($memory mb)"
	560	fi
[312]	561
	562	if [[ $remote_host = ibmh ]]
[1]	563	then
[312]	564	data_limit=""
	565	network_to_use=""
	566	class=""
	567	environment=""
[1]	568	else
[312]	569	class="# @ class = $queue"
	570	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	571	if [[ $queue = cdev ]]
[1]	572	then
[312]	573	data_limit="# @ data_limit = 1.76gb"
	574	network_to_use="# @ network.mpi = sn_all,shared,ip"
[1]	575	else
[312]	576	if [[ $remote_host = ibms ]]
	577	then
	578	network_to_use="# @ network.mpi = csss,shared,us"
	579	elif [[ $remote_host = ibmy ]]
	580	then
	581	network_to_use=""
	582	else
	583	network_to_use="# @ network.mpi = sn_all,shared,us"
	584	data_limit="# @ data_limit = 1.76gb"
	585	fi
[1]	586	fi
	587	fi
	588
	589	cat > $job_to_send << %%END%%
	590	#!/bin/ksh
[312]	591	# @ shell = /bin/ksh
[1]	592
	593	# @ job_type = parallel
	594	# @ job_name = palm
	595	# @ wall_clock_limit = ${timestring},$timestring
	596	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	597	# @ output = $remote_dayfile
	598	# @ error = $remote_dayfile
	599	# @ image_size = 50
[312]	600	$class
	601	$environment
[1]	602	$network_to_use
	603	$data_limit
	604	$notification
	605
	606	%%END%%
	607
	608	if (( nodes > 0 ))
	609	then
	610
	611	cat >> $job_to_send << %%END%%
	612	# @ node = $nodes
	613	# @ tasks_per_node = $tasks_per_node
	614	# @ node_usage = $node_usage
	615	# @ queue
	616
	617	%%END%%
	618
	619	else
	620
	621	if [[ $remote_host != ibmy ]]
	622	then
	623
	624	cat >> $job_to_send << %%END%%
	625	# @ blocking = unlimited
	626	# @ total_tasks = $numprocs
	627	# @ node_usage = $node_usage
	628	# @ queue
	629
	630	%%END%%
	631
	632	else
	633
	634	cat >> $job_to_send << %%END%%
	635	# @ node = 1
	636	# @ total_tasks = $numprocs
	637	# @ queue
	638
	639	%%END%%
	640
	641	fi
	642
	643	fi
	644
	645	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	646	then
	647
	648	cat > $job_to_send << %%END%%
	649	#!/bin/ksh
	650
	651	# @ job_type = serial
	652	# @ node_usage = $node_usage
	653	# @ job_name = palm
	654	# @ wall_clock_limit = ${timestring},$timestring
	655	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	656	# @ output = $remote_dayfile
	657	# @ error = $remote_dayfile
[312]	658	$class
[1]	659	$notification
	660
	661	# @ queue
	662
	663	%%END%%
	664
[108]	665	elif [[ $remote_host = lcfimm ]]
	666	then
	667
	668	if [[ $numprocs != 0 ]]
	669	then
	670	cat > $job_to_send << %%END%%
	671	#!/bin/ksh
	672	#PBS -N $job_name
[206]	673	#PBS -A geofysisk
[108]	674	#PBS -l walltime=$timestring
	675	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
	676	#PBS -l pmem=${memory}mb
	677	#PBS -m abe
[206]	678	#PBS -M bjorn.maronga@student.uib.no
[108]	679	#PBS -o $remote_dayfile
	680	#PBS -j oe
	681	mpd &
	682
	683	%%END%%
	684
[168]	685	else
	686	cat > $job_to_send << %%END%%
	687	#!/bin/ksh
	688	#PBS -N $job_name
[206]	689	#PBS -A geofysisk
[168]	690	#PBS -l walltime=$timestring
	691	#PBS -l ncpus=1
	692	#PBS -l pmem=${memory}mb
	693	#PBS -m abe
[206]	694	#PBS -M bjorn.maronga@student.uib.no
[168]	695	#PBS -o $remote_dayfile
	696	#PBS -j oe
	697
	698	%%END%%
	699
	700	fi
	701
[181]	702	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
[179]	703	then
	704
[191]	705	if [[ $queue = testq ]]
	706	then
	707	queue_directive="#PBS -q $queue"
	708	else
	709	queue_directive=""
	710	fi
	711
[179]	712	if [[ $numprocs != 0 ]]
	713	then
	714	cat > $job_to_send << %%END%%
[181]	715	#!/bin/bash
[179]	716	#PBS -N $job_name
	717	#PBS -l walltime=$timestring
[181]	718	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
	719	#PBS -l feature=$sgi_feature
	720	#PBS -l naccesspolicy=$node_usage
[179]	721	#PBS -o $remote_dayfile
	722	#PBS -j oe
[191]	723	$queue_directive
[179]	724
	725	. /usr/share/modules/init/bash
[221]	726	module load $mpilib
[196]	727	# module load mvapich2/1.2rc1-intel
	728	# module load mvapich2
[179]	729	module load netcdf
[181]	730	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
	731	echo ld_library_path=\$LD_LIBRARY_PATH
[179]	732
	733	%%END%%
	734
	735	else
	736	cat > $job_to_send << %%END%%
	737	#PBS -S /bin/bash
	738	#PBS -N $job_name
	739	#PBS -l walltime=$timestring
	740	#PBS -l ncpus=1
	741	#PBS -o $remote_dayfile
	742	#PBS -e $remote_dayfile
	743
	744	%%END%%
	745
	746	fi
	747
[164]	748	elif [[ $remote_host = lcxt4 ]]
	749	then
	750
	751	if [[ $numprocs != 0 ]]
	752	then
	753	cat > $job_to_send << %%END%%
	754	#!/bin/ksh
	755	#PBS -S /bin/ksh
	756	#PBS -N $job_name
[206]	757	#PBS -A geofysisk
	758	#PBS -j oe
[164]	759	#PBS -l walltime=$timestring
	760	#PBS -l mppwidth=${numprocs}
	761	#PBS -l mppnppn=${tasks_per_node}
	762	#PBS -m abe
[206]	763	#PBS -M bjorn.maronga@student.uib.no
[164]	764	#PBS -o $remote_dayfile
	765
	766	%%END%%
	767
[108]	768	else
	769	cat > $job_to_send << %%END%%
	770	#!/bin/ksh
[168]	771	#PBS -S /bin/ksh
[108]	772	#PBS -N $job_name
[206]	773	#PBS -A geofysisk
	774	#PBS -j oe
[108]	775	#PBS -l walltime=$timestring
	776	#PBS -l ncpus=1
	777	#PBS -l pmem=${memory}mb
	778	#PBS -m abe
[206]	779	#PBS -M bjorn.maronga@student.uib.no
[108]	780	#PBS -o $remote_dayfile
	781
	782	%%END%%
	783
	784	fi
	785
[1]	786	elif [[ $remote_host = nech ]]
	787	then
	788
	789	if (( nodes > 1 ))
	790	then
	791	cat > $job_to_send << %%END%%
	792	#!/bin/ksh
	793	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	794	#PBS -l ${qsubmem}=${Memory}gb
	795	#PBS -b $nodes
	796	#PBS -o $remote_dayfile
	797	#PBS -N palm
	798	#PBS -j o
	799	#PBS -T mpisx
	800
	801	%%END%%
	802
	803	elif [[ $numprocs != 0 ]]
	804	then
	805	cat > $job_to_send << %%END%%
	806	#!/bin/ksh
	807	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	808	#PBS -l ${qsubmem}=${Memory}gb
	809	#PBS -o $remote_dayfile
	810	#PBS -N palm
	811	#PBS -j o
	812
	813	%%END%%
	814
	815	else
	816	cat > $job_to_send << %%END%%
	817	#!/bin/ksh
	818	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	819	#PBS -o $remote_dayfile
	820	#PBS -j o
	821
	822	%%END%%
	823
	824	fi
	825
[251]	826	elif [[ $remote_host = necriam ]]
[35]	827	then
	828
	829	if (( nodes > 1 ))
	830	then
	831	cat > $job_to_send << %%END%%
	832	#!/bin/ksh
	833	#PBS -b $nodes
	834	#PBS -o $remote_dayfile
[253]	835	#PBS -N $job_name
[35]	836	#PBS -j o
[251]	837	#PBS -v MPIPROGINV=YES
[35]	838
	839	%%END%%
	840
	841	elif [[ $numprocs != 0 ]]
	842	then
	843	cat > $job_to_send << %%END%%
	844	#!/bin/ksh
	845	#PBS -q ${queue}
	846	#PBS -o $remote_dayfile
[253]	847	#PBS -N $job_name
[35]	848	#PBS -j o
[251]	849	#PBS -v MPIPROGINV=YES
[35]	850
	851	%%END%%
	852
	853	else
	854	cat > $job_to_send << %%END%%
	855	#!/bin/ksh
	856	#PBS -o $remote_dayfile
	857	#PBS -j o
	858
	859	%%END%%
	860
	861	fi
	862
[1]	863	elif [[ $remote_host = lctit ]]
	864	then
	865	cat > $job_to_send << %%END%%
	866	#!/bin/bash
	867	#$ -S /bin/bash
	868	cd $job_catalog
[69]	869	export PATH=$PALM_BIN:\$PATH
	870	echo \$PATH
[1]	871
	872	%%END%%
	873
	874	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
[118]	875	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
[122]	876	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
[1]	877
	878	else
	879
	880	cat > $job_to_send << %%END%%
	881	# @\$-q ${queue}
	882	# @\$-l${qsubtime} $timestring
	883	# @\$-l${qsubmem} ${memory}mb
	884	# @\$-o $remote_dayfile
	885	# @\$-eo
	886
	887	%%END%%
	888
	889	fi
	890
	891
	892	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	893	# BEI EXIT VERANLASSEN
	894	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	895	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	896	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	897	then
	898	echo "set +vx" >> $job_to_send
	899	echo "trap '" >> $job_to_send
	900	echo "set +vx" >> $job_to_send
[181]	901	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]	902	then
	903	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
	904	then
	905	return_queue=c1
	906	elif [[ $remote_host = ibms ]]
	907	then
	908	return_queue=p_normal
	909	elif [[ $remote_host = ibmy ]]
	910	then
	911	return_queue=serial
[325]	912	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
	913	then
[326]	914	return_queue=serialq
[1]	915	elif [[ $remote_host = lctit ]]
	916	then
[118]	917	return_queue=sla3
[251]	918	elif [[ $remote_host = necriam ]]
[1]	919	then
[253]	920	return_queue=SP
[1]	921	else
	922	return_queue=unknown
	923	fi
	924
	925	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	926	then
	927
	928	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	929	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	930	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	931	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	932	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	933	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	934	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
[312]	935	if [[ $host != "ibmh" ]]
	936	then
	937	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	938	fi
[1]	939	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	940	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	941
	942	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	943	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	944
	945	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	946	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
	947	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	948
	949	elif [[ $remote_host = nech ]]
	950	then
	951	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	952	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	953	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	954	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	955	echo "#PBS -j o" >> $job_to_send
	956	echo " " >> $job_to_send
	957	echo "set -x" >> $job_to_send
	958	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	959	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	960	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	961	echo "%%END%%" >> $job_to_send
	962
[251]	963	elif [[ $remote_host = necriam ]]
[35]	964	then
	965	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	966	echo "#PBS -q $return_queue" >> $job_to_send
	967	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	968	echo "#PBS -j o" >> $job_to_send
	969	echo " " >> $job_to_send
	970	echo "set -x" >> $job_to_send
	971	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	972	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	973	echo "%%END%%" >> $job_to_send
	974
[181]	975	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	976	then
	977	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	978	echo "#!/bin/bash" >> $job_to_send
	979	echo "#PBS -N job_protocol_transfer" >> $job_to_send
[325]	980	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[181]	981	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
	982	echo "#PBS -l feature=xe" >> $job_to_send
	983	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	984	echo "#PBS -j oe" >> $job_to_send
	985	echo " " >> $job_to_send
	986	echo ". /usr/share/modules/init/bash" >> $job_to_send
	987	echo "set -x" >> $job_to_send
	988	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	989	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	990	echo "%%END%%" >> $job_to_send
	991
[1]	992	elif [[ $remote_host = lctit ]]
	993	then
	994	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	995	echo "set -x" >> $job_to_send
	996	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
	997	echo "cd $job_catalog" >> $job_to_send
	998	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	999	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1000	echo "%%END%%" >> $job_to_send
	1001
	1002	else
	1003
	1004	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1005	echo "# @\\\$-q $return_queue" >> $job_to_send
	1006	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1007	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1008	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1009	then
	1010	echo "# @\$-l mpp_p=0" >> $job_to_send
	1011	fi
	1012	echo '# @\$-lF 10mb' >> $job_to_send
	1013	# echo '# @\$-o /dev/null' >> $job_to_send
	1014	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1015	echo '# @\\\$-eo' >> $job_to_send
	1016	echo " " >> $job_to_send
	1017	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1018	then
	1019	echo "set +vx" >> $job_to_send
	1020	echo ". .profile" >> $job_to_send
	1021	fi
	1022	echo "set -x" >> $job_to_send
	1023	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	1024	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1025	echo "%%END%%" >> $job_to_send
	1026	fi
	1027
	1028	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1029	then
	1030	echo "llsubmit scpjob.$kennung" >> $job_to_send
[181]	1031	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1032	then
	1033	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1034	echo "msub scpjob.$kennung" >> $job_to_send
[1]	1035	elif [[ $remote_host = lctit ]]
	1036	then
	1037	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1038	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
	1039	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
	1040	echo "sleep 10" >> $job_to_send
	1041	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1042	then
	1043	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	1044	elif [[ $remote_host = t3es ]]
	1045	then
	1046	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
	1047	else
	1048	echo "qsub scpjob.$kennung" >> $job_to_send
	1049	fi
	1050	echo "rm scpjob.$kennung" >> $job_to_send
	1051	if [[ $remote_host = nech ]]
	1052	then
	1053	echo "cd -" >> $job_to_send
	1054	fi
	1055	else
	1056	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1057	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1058	fi
	1059	echo "set -x" >> $job_to_send
	1060	echo " ' exit" >> $job_to_send
	1061	echo "set -x" >> $job_to_send
	1062	fi
	1063
	1064
	1065
	1066	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	1067	cat $file_to_send >> $job_to_send
[69]	1068
[1]	1069	if [[ $remote_host = ibm ]]
	1070	then
	1071	echo " " >> $job_to_send
	1072	echo "exit" >> $job_to_send
	1073	fi
[69]	1074	if [[ $remote_host = lctit ]]
	1075	then
	1076	echo " " >> $job_to_send
	1077	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1078	fi
[1]	1079
	1080
	1081
	1082	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	1083	if [[ -z $remote_user ]]
	1084	then
	1085	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	1086	then
	1087	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	1088	fi
	1089	fi
	1090
	1091
	1092
	1093	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	1094	if [[ $no_submit = false ]]
	1095	then
	1096	if [[ $remote_host != $local_host ]]
	1097	then
	1098	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	1099	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	1100	then
[82]	1101	job_catalog_save=$job_catalog
	1102	job_catalog=job_queue
	1103	elif [[ $remote_host = nech ]]
	1104	then
	1105	job_catalog_save=$job_catalog
	1106	job_catalog=/hpf/b/${remote_user}/job_queue
	1107	fi
	1108	if [[ $local_host = decalpha ]]
	1109	then
	1110	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	1111	# PASSWORT AB
	1112	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	1113	elif [[ $remote_host = nech ]]
	1114	then
	1115	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	1116	# TRANSFERIERT WERDEN
	1117	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1118	else
[82]	1119	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1120	fi
[82]	1121	if [[ $? = 1 ]]
	1122	then
	1123	locat=scp; exit
	1124	fi
	1125	if [[ $remote_host = ibms ]]
	1126	then
	1127	job_catalog=$job_catalog_save
	1128	fi
[1]	1129	[[ $verify = true ]] && printf "\n >>> finished\n"
	1130	else
	1131	eval job_catalog=$job_catalog
	1132	cp $job_to_send ${job_catalog}/$job_on_remhost
	1133	fi
	1134
	1135
	1136
	1137	# NQS- BZW. LOADLEVELER-JOB STARTEN
	1138	if [[ $remote_host != $local_host ]]
	1139	then
	1140	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[82]	1141	if [[ $remote_host != lctit ]]
[1]	1142	then
[191]	1143	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[1]	1144	else
[206]	1145	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
[82]	1146	# MIT SSH, DESHALB AUFRUF PER PIPE
	1147	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
	1148	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	1149	echo "first try" > send_protocol
	1150	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
	1151	do
	1152	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
	1153	then
	1154	printf "\n trying to submit job to TIT cluster..."
	1155	else
	1156	printf "\n +++ failed ... trying to submit again ..."
	1157	fi
	1158	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
	1159	cat send_protocol
	1160	done
	1161	sleep 10
	1162	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	1163	echo "$submcom $job_on_remhost"
	1164	rm send_protocol
	1165	fi
[1]	1166
	1167	[[ $verify = true ]] && printf " >>> o.k.\n"
	1168	else
	1169	cd $job_catalog
[210]	1170	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
[1]	1171	then
	1172	eval $submcom $job_on_remhost
[164]	1173	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 ]]
[108]	1174	then
	1175	eval $submcom $job_on_remhost
	1176	echo "$submcom $job_on_remhost"
	1177	chmod u+x $job_on_remhost
[292]	1178	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
	1179	then
	1180	eval $submcom $job_on_remhost
[1]	1181	elif [[ $local_host = nech ]]
	1182	then
	1183	if [[ $queue = default ]]
	1184	then
	1185	eval $submcom $job_on_remhost
	1186	else
	1187	eval $submcom -q $queue $job_on_remhost
	1188	fi
	1189	else
	1190	qsub $job_on_remhost
	1191	fi
[69]	1192	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
	1193	[[ $local_host != lctit ]] && rm $job_on_remhost
[1]	1194	cd - > /dev/null
	1195	fi
	1196	fi
	1197
	1198
	1199
	1200	# ABSCHLUSSARBEITEN
	1201
	1202	if [[ $no_submit = false ]]
	1203	then
	1204	rm $job_to_send
	1205	fi
	1206	[[ $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 |