Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 269

Last change on this file since 269 was 259, checked in by raasch, 16 years ago
subjob update for NEC-SX9 at RIAM
Property svn:keywords set to `Id Rev`
File size: 41.3 KB

Rev	Line
[1]	1	#!/bin/ksh
[82]	2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
[169]	3	# $Id: subjob 259 2009-03-16 00:21:34Z weinreis $
[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
[1]	98
	99
[206]	100
[1]	101	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
	102	delete_dayfile=false
[122]	103	group_number=none
[1]	104	locat=normal
	105	no_default_queue=none
	106	no_submit=false
	107	job_catalog="~/job_queue"
	108	job_name=none
	109	local_user=$LOGNAME
	110	node_usage=shared
	111	numprocs=1
	112	punkte="..........................................................."
	113	submcom=qsub
	114	queue=default
	115	remote_host=none
	116	remote_user=""
	117	verify=true
	118
	119	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
	120	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
	121	typeset -L20 spalte1
	122	typeset -R40 spalte2
	123	typeset -L60 spalte3
	124
	125
	126
	127	# FEHLERBEHANDLUNG
	128	# BEI EXIT:
	129	trap 'if [[ $locat != normal ]]
	130	then
	131	case $locat in
	132	(option) printf "\n --> available optios can be displayed"
	133	printf " by typing:"
	134	printf "\n \"subjob ?\" \n";;
	135	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	136	(*) printf "\n +++ unknown error"
	137	printf "\n please inform S. Raasch!\n"
	138	esac
	139	[[ -f $job_to_send ]] && rm $job_to_send
	140	printf "\n\n+++ SUBJOB killed \n\n"
	141	fi' exit
	142
	143
	144	# BEI TERMINAL-BREAK:
	145	trap '[[ -f $job_to_send ]] && rm $job_to_send
	146	printf "\n\n+++ SUBJOB killed \n\n"
	147	exit
	148	' 2
	149
	150
	151
	152
	153	# LOKALEN HOSTNAMEN FESTSTELLEN
	154	local_host=$(hostname)
	155
	156
	157
	158	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
	159	# UEBERHAUPT ZULAESSIG IST
	160	case $local_host in
	161	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[102]	162	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	163	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[210]	164	(bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
[1]	165	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
[12]	166	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[196]	167	(b01*) local_addres=130.73.232.102; local_host=lcsgib;;
[108]	168	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
[1]	169	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	170	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[108]	171	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
[101]	172	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[1]	173	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	174	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	175	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	176	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[164]	177	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
[210]	178	(hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
	179	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
[1]	180	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
[181]	181	(h01*) local_addres=130.75.4.102; local_host=lcsgih;;
[5]	182	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
	183	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
	184	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[164]	185	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
[1]	186	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	187	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	188	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	189	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[5]	190	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[1]	191	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
	192	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
[259]	193	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
[181]	194	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
[196]	195	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
[1]	196	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
	197	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
	198	(*) printf "\n +++ \"$local_host\" unknown";
	199	printf "\n please inform S. Raasch!";
	200	locat=parameter; exit;;
	201	esac
	202
	203
	204
	205	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	206	remote_host=$local_host
	207
	208
	209
	210
	211	# PROZEDUROPTIONEN EINLESEN
[122]	212	while getopts :c:dDg:h:m:n:N:O:q:t:T:u:vX: option
[1]	213	do
	214	case $option in
	215	(c) job_catalog=$OPTARG;;
	216	(d) delete_dayfile=true;;
	217	(D) no_submit=true;;
[125]	218	(g) group_number=$OPTARG;;
[1]	219	(h) remote_host=$OPTARG;;
	220	(m) memory=$OPTARG;;
	221	(n) job_name=$OPTARG;;
	222	(N) node_usage=$OPTARG;;
	223	(O) threads_per_task=$OPTARG;;
	224	(q) no_default_queue=$OPTARG;;
	225	(t) cputime=$OPTARG;;
	226	(T) tasks_per_node=$OPTARG;;
	227	(u) remote_user=$OPTARG;;
	228	(v) verify=false;;
	229	(X) numprocs=$OPTARG;;
	230	(\?) printf "\n +++ Option $OPTARG unknown \n";
	231	locat=option; exit;;
	232	esac
	233	done
	234
	235
	236	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	237	shift OPTIND-1; file_to_send=$1
	238
	239
	240	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	241	if [ "$1" = "?" ]
	242	then
	243	(printf "\n *** subjob can be called as follows:\n"
	244	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	245	printf "\n Description of available options:\n"
	246	printf "\n Option Description Default-Value"
	247	printf "\n -c job-input- and output-catalog ~/job_queue"
	248	printf "\n -d no job-protocol will be created ---"
	249	printf "\n -D only the job-file will be created ---"
	250	printf "\n -h execution host, available hosts: $remote_host"
[5]	251	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
[251]	252	printf "\n lctit, nech, necriam, unics"
[1]	253	printf "\n -m memory demand per process in MByte ---"
	254	printf "\n -n jobname <jobdatei>"
	255	printf "\n -O threads per task (for OpenMP usage) 1"
	256	printf "\n -q job-queue to be used default"
	257	printf "\n -t allowed cpu-time in seconds ---"
	258	printf "\n -T tasks per node (on parallel hosts) ---"
	259	printf "\n -u username on execution host from .netrc"
	260	printf "\n -v no prompt for confirmation ---"
	261	printf "\n -X # of processors (on parallel hosts) 1"
	262	printf "\n "
	263	printf "\n The only possible positional parameter is <jobfile>:"
	264	printf "\n The complete NQS-job must be provided here."
	265	printf "\n <jobfile>=? creates this outline\n\n") \| more
	266	exit
	267	fi
	268
	269
	270
	271	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	272	if [[ "$file_to_send" = "" ]]
	273	then
	274	printf "\n +++ job-file missing"
	275	locat=parameter; exit
	276	else
	277	if [[ -f $file_to_send ]]
	278	then
	279	true
	280	else
	281	printf "\n +++ job-file: "
	282	printf "\n $file_to_send"
	283	printf "\n does not exist"
	284	locat=parameter; exit
	285	fi
	286	fi
	287
	288
	289
	290	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	291	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	292	if [[ $job_name = none ]]
	293	then
	294	job_name=$file_to_send
	295	fi
	296	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	297	then
	298	printf "\n +++ job-file name: "
	299	printf "\n $job_name"
	300	printf "\n must not contain \"/\"-characters"
	301	locat=parameter; exit
	302	fi
	303
	304
	305
	306
	307	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	308	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	309	if [[ $remote_host = none ]]
	310	then
	311	printf "\n +++ host missing"
	312	locat=option; exit
	313	else
	314	case $remote_host in
[251]	315	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	316	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	317	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	318	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	319	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	320	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
	321	(lcsgib) queue=smallq; remote_addres=130.73.232.102; submcom=/opt/moab/bin/msub;;
	322	(lcsgih) queue=smallq; remote_addres=130.75.4.103; submcom=/opt/moab/bin/msub;;
	323	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
	324	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
	325	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	326	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
	327	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	328	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	329	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	330	locat=parameter; exit;;
[1]	331	esac
	332	fi
	333
	334
	335	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	336	if [[ $no_default_queue != none ]]
	337	then
	338	error=false
	339	ndq=$no_default_queue
	340	case $remote_host in
	341	(ibm) case $ndq in
	342	(p690_express\|p690_standard\|p690_long) error=false;;
	343	(*) error=true;;
	344	esac;;
	345	(ibmb) case $ndq in
	346	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
	347	(*) error=true;;
	348	esac;;
	349	(ibmh) case $ndq in
[113]	350	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|csoloh3h\|csoloh6h\|csoloh12h\|csoloh1d\|csoloh2d\|cspec) error=false;;
[1]	351	(*) error=true;;
	352	esac;;
	353	(ibms) case $ndq in
	354	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	355	(*) error=true;;
	356	esac;;
	357	(ibmy) case $ndq in
	358	(parallel) error=false;;
	359	(*) error=true;;
	360	esac;;
[191]	361	(lcsgib\|lcsgih) case $ndq in
	362	(testq\|smallq\|bigq) error=false;;
	363	(*) error=true;;
	364	esac;;
[1]	365	(lctit) case $ndq in
[129]	366	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
[1]	367	(*) error=true;;
	368	esac;;
	369	(t3eb) case $ndq in
	370	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	371	(*) error=true;;
	372	esac;;
[251]	373	(necriam) case $ndq in
[253]	374	(SP\|SS\|P6) error=false;;
[251]	375	(*) error=true;;
	376	esac;;
[1]	377	(t3eh) case $ndq in
	378	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	379	(*) error=true;;
	380	esac;;
	381	(t3ej2\|t3ej5) case $ndq in
	382	(low\|normal\|high) error=false;;
	383	(*) error=true;;
	384	esac;;
	385	(t3es) case $ndq in
	386	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	387	(*) error=true;;
	388	esac;;
	389	(unics) case $ndq in
	390	(unics\|ht) error=false;;
	391	(*) error=true;;
	392	esac;;
	393	esac
	394	if [[ $error = true ]]
	395	then
	396	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	397	locat=parameter; exit
	398	else
	399	queue=$no_default_queue
	400	fi
	401	fi
	402
	403
	404
	405	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
	406	if [[ $node_usage != shared && $queue = cdev ]]
	407	then
	408	node_usage=shared
	409	fi
	410
	411
	412
	413	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	414	# AUFTEILEN
	415	done=false
	416	while [[ $done = false ]]
	417	do
	418	if (( $cputime <= 0 ))
	419	then
	420	printf "\n +++ wrong cpu-time or cpu-time missing"
	421	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	422	printf "\n >>> "
	423	read cputime 1>/dev/null 2>&1
	424	else
	425	done=true
	426	fi
	427	done
	428	if [[ $remote_host = nech ]]
	429	then
	430	if (( tasks_per_node != 0 ))
	431	then
	432	(( cputime = cputime * tasks_per_node ))
	433	elif [[ $numprocs != 0 ]]
	434	then
	435	(( cputime = cputime * numprocs ))
	436	fi
	437	fi
	438	(( stunden = cputime / 3600 ))
	439	(( resttime = cputime - stunden * 3600 ))
	440	(( minuten = resttime / 60 ))
	441	(( sekunden = resttime - minuten * 60 ))
	442	timestring=${stunden}:${minuten}:${sekunden}
	443
	444
	445
	446	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	447	done=false
	448	while [[ $done = false ]]
	449	do
	450	if (( memory <= 0 ))
	451	then
	452	printf "\n +++ wrong memory demand or memory demand missing"
	453	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	454	printf "\n >>> "
	455	read memory 1>/dev/null 2>&1
	456	else
	457	done=true
	458	fi
	459	done
	460
[251]	461	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
[1]	462	then
	463	if (( tasks_per_node != 0 ))
	464	then
	465	(( Memory = memory * tasks_per_node / 1000 ))
	466	elif [[ $numprocs != 0 ]]
	467	then
	468	(( Memory = memory * numprocs / 1000 ))
	469	else
	470	(( Memory = memory / 1000 ))
	471	fi
	472	fi
	473
	474
	475	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	476	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	477	then
	478	(( memory = memory * threads_per_task ))
	479	fi
	480
	481
	482	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	483	if (( tasks_per_node != 0 ))
	484	then
	485	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
	486	fi
	487
	488
	489
	490	# HEADER-AUSGABE
	491	if [[ $verify = true ]]
	492	then
	493	printf "\n\n"
	494	printf "#--------------------------------------------------------------# \n"
	495	spalte1=SUBJOB;spalte2=$(date)
	496	printf "\| $spalte1$spalte2 \| \n"
	497	printf "\| \| \n"
	498	printf "\| values of parameters/options: \| \n"
	499	spalte1=local_host$punkte; spalte2=$punkte$local_host
	500	printf "\| $spalte1$spalte2 \| \n"
	501	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	502	printf "\| $spalte1$spalte2 \| \n"
	503	spalte1=queue$punkte; spalte2=$punkte$queue
	504	printf "\| $spalte1$spalte2 \| \n"
	505	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	506	printf "\| $spalte1$spalte2 \| \n"
	507	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	508	printf "\| $spalte1$spalte2 \| \n"
	509	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	510	printf "\| $spalte1$spalte2 \| \n"
	511	printf "#--------------------------------------------------------------# \n\n"
	512
	513
	514	# KONTROLLABFRAGE, OB ALLES O.K.
	515	antwort="dummy"
	516	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	517	do
	518	read antwort?" >>> continue (y/n) ? "
	519	done
	520	if [[ $antwort = n \|\| $antwort = N ]]
	521	then
	522	locat=verify; exit
	523	fi
	524	printf "\n"
	525	fi
	526
	527
	528
	529	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	530	kennung=$RANDOM
	531	job_on_remhost=${job_name}_${kennung}_$local_host
	532	job_to_send=job_to_send_$kennung
	533	if [[ $delete_dayfile = false ]]
	534	then
	535	remote_dayfile=${local_host}_${job_name}_result_$kennung
	536	local_dayfile=${remote_host}_${job_name}
	537	else
	538	remote_dayfile=/dev/null
	539	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	540	then
	541	notification='# @ notification = never'
	542	fi
	543	fi
	544
	545
	546
	547	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
	548	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	549	then
	550
	551	if [[ $remote_host = ibmy ]]
	552	then
	553	consumable_memory=""
	554	else
	555	consumable_memory="ConsumableMemory($memory mb)"
	556	fi
	557	if [[ $queue = cdev ]]
	558	then
	559	data_limit="# @ data_limit = 1.76gb"
	560	network_to_use="# @ network.mpi = sn_all,shared,ip"
	561	else
	562	if [[ $remote_host = ibms ]]
	563	then
	564	network_to_use="# @ network.mpi = csss,shared,us"
	565	elif [[ $remote_host = ibmy ]]
	566	then
	567	network_to_use=""
	568	else
	569	network_to_use="# @ network.mpi = sn_all,shared,us"
	570	data_limit="# @ data_limit = 1.76gb"
	571	fi
	572	fi
	573
	574	cat > $job_to_send << %%END%%
	575	#!/bin/ksh
	576
	577	# @ job_type = parallel
	578	# @ job_name = palm
	579	# @ wall_clock_limit = ${timestring},$timestring
	580	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	581	# @ output = $remote_dayfile
	582	# @ error = $remote_dayfile
	583	# @ image_size = 50
	584	# @ class = $queue
	585	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
	586	$network_to_use
	587	$data_limit
	588	$notification
	589
	590	%%END%%
	591
	592	if (( nodes > 0 ))
	593	then
	594
	595	cat >> $job_to_send << %%END%%
	596	# @ node = $nodes
	597	# @ tasks_per_node = $tasks_per_node
	598	# @ node_usage = $node_usage
	599	# @ queue
	600
	601	%%END%%
	602
	603	else
	604
	605	if [[ $remote_host != ibmy ]]
	606	then
	607
	608	cat >> $job_to_send << %%END%%
	609	# @ blocking = unlimited
	610	# @ total_tasks = $numprocs
	611	# @ node_usage = $node_usage
	612	# @ queue
	613
	614	%%END%%
	615
	616	else
	617
	618	cat >> $job_to_send << %%END%%
	619	# @ node = 1
	620	# @ total_tasks = $numprocs
	621	# @ queue
	622
	623	%%END%%
	624
	625	fi
	626
	627	fi
	628
	629	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	630	then
	631
	632	cat > $job_to_send << %%END%%
	633	#!/bin/ksh
	634
	635	# @ job_type = serial
	636	# @ node_usage = $node_usage
	637	# @ job_name = palm
	638	# @ wall_clock_limit = ${timestring},$timestring
	639	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	640	# @ output = $remote_dayfile
	641	# @ error = $remote_dayfile
	642	# @ class = $queue
	643	$notification
	644
	645	# @ queue
	646
	647	%%END%%
	648
[108]	649	elif [[ $remote_host = lcfimm ]]
	650	then
	651
	652	if [[ $numprocs != 0 ]]
	653	then
	654	cat > $job_to_send << %%END%%
	655	#!/bin/ksh
	656	#PBS -N $job_name
[206]	657	#PBS -A geofysisk
[108]	658	#PBS -l walltime=$timestring
	659	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
	660	#PBS -l pmem=${memory}mb
	661	#PBS -m abe
[206]	662	#PBS -M bjorn.maronga@student.uib.no
[108]	663	#PBS -o $remote_dayfile
	664	#PBS -j oe
	665	mpd &
	666
	667	%%END%%
	668
[168]	669	else
	670	cat > $job_to_send << %%END%%
	671	#!/bin/ksh
	672	#PBS -N $job_name
[206]	673	#PBS -A geofysisk
[168]	674	#PBS -l walltime=$timestring
	675	#PBS -l ncpus=1
	676	#PBS -l pmem=${memory}mb
	677	#PBS -m abe
[206]	678	#PBS -M bjorn.maronga@student.uib.no
[168]	679	#PBS -o $remote_dayfile
	680	#PBS -j oe
	681
	682	%%END%%
	683
	684	fi
	685
[181]	686	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
[179]	687	then
	688
[191]	689	if [[ $queue = testq ]]
	690	then
	691	queue_directive="#PBS -q $queue"
	692	else
	693	queue_directive=""
	694	fi
	695
[179]	696	if [[ $numprocs != 0 ]]
	697	then
	698	cat > $job_to_send << %%END%%
[181]	699	#!/bin/bash
[179]	700	#PBS -N $job_name
	701	#PBS -l walltime=$timestring
[181]	702	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
	703	#PBS -l feature=$sgi_feature
	704	#PBS -l naccesspolicy=$node_usage
[179]	705	#PBS -o $remote_dayfile
	706	#PBS -j oe
[191]	707	$queue_directive
[179]	708
	709	. /usr/share/modules/init/bash
[221]	710	module load $mpilib
[196]	711	# module load mvapich2/1.2rc1-intel
	712	# module load mvapich2
[179]	713	module load netcdf
[181]	714	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
	715	echo ld_library_path=\$LD_LIBRARY_PATH
[179]	716
	717	%%END%%
	718
	719	else
	720	cat > $job_to_send << %%END%%
	721	#PBS -S /bin/bash
	722	#PBS -N $job_name
	723	#PBS -l walltime=$timestring
	724	#PBS -l ncpus=1
	725	#PBS -o $remote_dayfile
	726	#PBS -e $remote_dayfile
	727
	728	%%END%%
	729
	730	fi
	731
[164]	732	elif [[ $remote_host = lcxt4 ]]
	733	then
	734
	735	if [[ $numprocs != 0 ]]
	736	then
	737	cat > $job_to_send << %%END%%
	738	#!/bin/ksh
	739	#PBS -S /bin/ksh
	740	#PBS -N $job_name
[206]	741	#PBS -A geofysisk
	742	#PBS -j oe
[164]	743	#PBS -l walltime=$timestring
	744	#PBS -l mppwidth=${numprocs}
	745	#PBS -l mppnppn=${tasks_per_node}
	746	#PBS -m abe
[206]	747	#PBS -M bjorn.maronga@student.uib.no
[164]	748	#PBS -o $remote_dayfile
	749
	750	%%END%%
	751
[108]	752	else
	753	cat > $job_to_send << %%END%%
	754	#!/bin/ksh
[168]	755	#PBS -S /bin/ksh
[108]	756	#PBS -N $job_name
[206]	757	#PBS -A geofysisk
	758	#PBS -j oe
[108]	759	#PBS -l walltime=$timestring
	760	#PBS -l ncpus=1
	761	#PBS -l pmem=${memory}mb
	762	#PBS -m abe
[206]	763	#PBS -M bjorn.maronga@student.uib.no
[108]	764	#PBS -o $remote_dayfile
	765
	766	%%END%%
	767
	768	fi
	769
[1]	770	elif [[ $remote_host = nech ]]
	771	then
	772
	773	if (( nodes > 1 ))
	774	then
	775	cat > $job_to_send << %%END%%
	776	#!/bin/ksh
	777	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	778	#PBS -l ${qsubmem}=${Memory}gb
	779	#PBS -b $nodes
	780	#PBS -o $remote_dayfile
	781	#PBS -N palm
	782	#PBS -j o
	783	#PBS -T mpisx
	784
	785	%%END%%
	786
	787	elif [[ $numprocs != 0 ]]
	788	then
	789	cat > $job_to_send << %%END%%
	790	#!/bin/ksh
	791	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	792	#PBS -l ${qsubmem}=${Memory}gb
	793	#PBS -o $remote_dayfile
	794	#PBS -N palm
	795	#PBS -j o
	796
	797	%%END%%
	798
	799	else
	800	cat > $job_to_send << %%END%%
	801	#!/bin/ksh
	802	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	803	#PBS -o $remote_dayfile
	804	#PBS -j o
	805
	806	%%END%%
	807
	808	fi
	809
[251]	810	elif [[ $remote_host = necriam ]]
[35]	811	then
	812
	813	if (( nodes > 1 ))
	814	then
	815	cat > $job_to_send << %%END%%
	816	#!/bin/ksh
	817	#PBS -b $nodes
	818	#PBS -o $remote_dayfile
[253]	819	#PBS -N $job_name
[35]	820	#PBS -j o
[251]	821	#PBS -v MPIPROGINV=YES
[35]	822
	823	%%END%%
	824
	825	elif [[ $numprocs != 0 ]]
	826	then
	827	cat > $job_to_send << %%END%%
	828	#!/bin/ksh
	829	#PBS -q ${queue}
	830	#PBS -o $remote_dayfile
[253]	831	#PBS -N $job_name
[35]	832	#PBS -j o
[251]	833	#PBS -v MPIPROGINV=YES
[35]	834
	835	%%END%%
	836
	837	else
	838	cat > $job_to_send << %%END%%
	839	#!/bin/ksh
	840	#PBS -o $remote_dayfile
	841	#PBS -j o
	842
	843	%%END%%
	844
	845	fi
	846
[1]	847	elif [[ $remote_host = lctit ]]
	848	then
	849	cat > $job_to_send << %%END%%
	850	#!/bin/bash
	851	#$ -S /bin/bash
	852	cd $job_catalog
[69]	853	export PATH=$PALM_BIN:\$PATH
	854	echo \$PATH
[1]	855
	856	%%END%%
	857
	858	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
[118]	859	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
[122]	860	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
[1]	861
	862	else
	863
	864	cat > $job_to_send << %%END%%
	865	# @\$-q ${queue}
	866	# @\$-l${qsubtime} $timestring
	867	# @\$-l${qsubmem} ${memory}mb
	868	# @\$-o $remote_dayfile
	869	# @\$-eo
	870
	871	%%END%%
	872
	873	fi
	874
	875
	876	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	877	# BEI EXIT VERANLASSEN
	878	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	879	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	880	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	881	then
	882	echo "set +vx" >> $job_to_send
	883	echo "trap '" >> $job_to_send
	884	echo "set +vx" >> $job_to_send
[181]	885	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]	886	then
	887	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
	888	then
	889	return_queue=c1
	890	elif [[ $remote_host = ibms ]]
	891	then
	892	return_queue=p_normal
	893	elif [[ $remote_host = ibmy ]]
	894	then
	895	return_queue=serial
	896	elif [[ $remote_host = lctit ]]
	897	then
[118]	898	return_queue=sla3
[251]	899	elif [[ $remote_host = necriam ]]
[1]	900	then
[253]	901	return_queue=SP
[1]	902	else
	903	return_queue=unknown
	904	fi
	905
	906	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	907	then
	908
	909	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	910	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	911	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	912	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	913	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	914	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	915	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	916	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	917	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	918	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	919
	920	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	921	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	922
	923	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	924	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
	925	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	926
	927	elif [[ $remote_host = nech ]]
	928	then
	929	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	930	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	931	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	932	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	933	echo "#PBS -j o" >> $job_to_send
	934	echo " " >> $job_to_send
	935	echo "set -x" >> $job_to_send
	936	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	937	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	938	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	939	echo "%%END%%" >> $job_to_send
	940
[251]	941	elif [[ $remote_host = necriam ]]
[35]	942	then
	943	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	944	echo "#PBS -q $return_queue" >> $job_to_send
	945	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	946	echo "#PBS -j o" >> $job_to_send
	947	echo " " >> $job_to_send
	948	echo "set -x" >> $job_to_send
	949	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	950	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	951	echo "%%END%%" >> $job_to_send
	952
[181]	953	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	954	then
	955	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	956	echo "#!/bin/bash" >> $job_to_send
	957	echo "#PBS -N job_protocol_transfer" >> $job_to_send
	958	echo "#PBS -l walltime=00:02:00" >> $job_to_send
	959	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
	960	echo "#PBS -l feature=xe" >> $job_to_send
	961	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	962	echo "#PBS -j oe" >> $job_to_send
	963	echo " " >> $job_to_send
	964	echo ". /usr/share/modules/init/bash" >> $job_to_send
	965	echo "set -x" >> $job_to_send
	966	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	967	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	968	echo "%%END%%" >> $job_to_send
	969
[1]	970	elif [[ $remote_host = lctit ]]
	971	then
	972	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	973	echo "set -x" >> $job_to_send
	974	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
	975	echo "cd $job_catalog" >> $job_to_send
	976	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	977	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	978	echo "%%END%%" >> $job_to_send
	979
	980	else
	981
	982	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	983	echo "# @\\\$-q $return_queue" >> $job_to_send
	984	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	985	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	986	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	987	then
	988	echo "# @\$-l mpp_p=0" >> $job_to_send
	989	fi
	990	echo '# @\$-lF 10mb' >> $job_to_send
	991	# echo '# @\$-o /dev/null' >> $job_to_send
	992	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	993	echo '# @\\\$-eo' >> $job_to_send
	994	echo " " >> $job_to_send
	995	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	996	then
	997	echo "set +vx" >> $job_to_send
	998	echo ". .profile" >> $job_to_send
	999	fi
	1000	echo "set -x" >> $job_to_send
	1001	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	1002	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1003	echo "%%END%%" >> $job_to_send
	1004	fi
	1005
	1006	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1007	then
	1008	echo "llsubmit scpjob.$kennung" >> $job_to_send
[181]	1009	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1010	then
	1011	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1012	echo "msub scpjob.$kennung" >> $job_to_send
[1]	1013	elif [[ $remote_host = lctit ]]
	1014	then
	1015	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1016	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
	1017	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
	1018	echo "sleep 10" >> $job_to_send
	1019	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1020	then
	1021	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	1022	elif [[ $remote_host = t3es ]]
	1023	then
	1024	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
	1025	else
	1026	echo "qsub scpjob.$kennung" >> $job_to_send
	1027	fi
	1028	echo "rm scpjob.$kennung" >> $job_to_send
	1029	if [[ $remote_host = nech ]]
	1030	then
	1031	echo "cd -" >> $job_to_send
	1032	fi
	1033	else
	1034	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1035	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1036	fi
	1037	echo "set -x" >> $job_to_send
	1038	echo " ' exit" >> $job_to_send
	1039	echo "set -x" >> $job_to_send
	1040	fi
	1041
	1042
	1043
	1044	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	1045	cat $file_to_send >> $job_to_send
[69]	1046
[1]	1047	if [[ $remote_host = ibm ]]
	1048	then
	1049	echo " " >> $job_to_send
	1050	echo "exit" >> $job_to_send
	1051	fi
[69]	1052	if [[ $remote_host = lctit ]]
	1053	then
	1054	echo " " >> $job_to_send
	1055	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1056	fi
[1]	1057
	1058
	1059
	1060	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	1061	if [[ -z $remote_user ]]
	1062	then
	1063	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	1064	then
	1065	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	1066	fi
	1067	fi
	1068
	1069
	1070
	1071	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	1072	if [[ $no_submit = false ]]
	1073	then
	1074	if [[ $remote_host != $local_host ]]
	1075	then
	1076	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	1077	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	1078	then
[82]	1079	job_catalog_save=$job_catalog
	1080	job_catalog=job_queue
	1081	elif [[ $remote_host = nech ]]
	1082	then
	1083	job_catalog_save=$job_catalog
	1084	job_catalog=/hpf/b/${remote_user}/job_queue
	1085	fi
	1086	if [[ $local_host = decalpha ]]
	1087	then
	1088	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	1089	# PASSWORT AB
	1090	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	1091	elif [[ $remote_host = nech ]]
	1092	then
	1093	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	1094	# TRANSFERIERT WERDEN
	1095	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1096	else
[82]	1097	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1098	fi
[82]	1099	if [[ $? = 1 ]]
	1100	then
	1101	locat=scp; exit
	1102	fi
	1103	if [[ $remote_host = ibms ]]
	1104	then
	1105	job_catalog=$job_catalog_save
	1106	fi
[1]	1107	[[ $verify = true ]] && printf "\n >>> finished\n"
	1108	else
	1109	eval job_catalog=$job_catalog
	1110	cp $job_to_send ${job_catalog}/$job_on_remhost
	1111	fi
	1112
	1113
	1114
	1115	# NQS- BZW. LOADLEVELER-JOB STARTEN
	1116	if [[ $remote_host != $local_host ]]
	1117	then
	1118	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[82]	1119	if [[ $remote_host != lctit ]]
[1]	1120	then
[191]	1121	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[1]	1122	else
[206]	1123	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
[82]	1124	# MIT SSH, DESHALB AUFRUF PER PIPE
	1125	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
	1126	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	1127	echo "first try" > send_protocol
	1128	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
	1129	do
	1130	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
	1131	then
	1132	printf "\n trying to submit job to TIT cluster..."
	1133	else
	1134	printf "\n +++ failed ... trying to submit again ..."
	1135	fi
	1136	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
	1137	cat send_protocol
	1138	done
	1139	sleep 10
	1140	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	1141	echo "$submcom $job_on_remhost"
	1142	rm send_protocol
	1143	fi
[1]	1144
	1145	[[ $verify = true ]] && printf " >>> o.k.\n"
	1146	else
	1147	cd $job_catalog
[210]	1148	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
[1]	1149	then
	1150	eval $submcom $job_on_remhost
[164]	1151	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 ]]
[108]	1152	then
	1153	eval $submcom $job_on_remhost
	1154	echo "$submcom $job_on_remhost"
	1155	chmod u+x $job_on_remhost
[1]	1156	elif [[ $local_host = nech ]]
	1157	then
	1158	if [[ $queue = default ]]
	1159	then
	1160	eval $submcom $job_on_remhost
	1161	else
	1162	eval $submcom -q $queue $job_on_remhost
	1163	fi
	1164	else
	1165	qsub $job_on_remhost
	1166	fi
[69]	1167	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
	1168	[[ $local_host != lctit ]] && rm $job_on_remhost
[1]	1169	cd - > /dev/null
	1170	fi
	1171	fi
	1172
	1173
	1174
	1175	# ABSCHLUSSARBEITEN
	1176
	1177	if [[ $no_submit = false ]]
	1178	then
	1179	rm $job_to_send
	1180	fi
	1181	[[ $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 |