Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 399

Last change on this file since 399 was 399, checked in by weinreis, 15 years ago
mrun and subjob modified for archiving and special1q
Property svn:keywords set to `Id Rev`
File size: 44.6 KB

Rev	Line
[1]	1	#!/bin/ksh
[82]	2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
[169]	3	# $Id: subjob 399 2009-10-20 13:23:02Z 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
[292]	98	# 16/04/09 - Marcus- Adjustments for lcsgib and lcsgih
[312]	99	# 21/04/09 - Siggi - adjustments for new IBM at DKRZ, which is now ibmh
[325]	100	# 18/05/09 - Siggi - Settings for serial jobs on lcsgi changed
[343]	101	# 24/06/09 - BjornM- adjustments for lcxt4 (loading modules manually)
[352]	102	# 08/07/09 - Siggi - option -e added (email notification on lcsgih/b)
[366]	103	# 20/07/09 - Siggi - On lcsgi, jobs for returning the job protocol are
	104	# now run on the data nodes (feature=data)
[367]	105	# 25/08/09 - BjornM- adapted for lck
[372]	106	# 26/08/09 - Marcus- caurus admitted; optional qos feature hiprio on lcsgi
[376]	107	# 03/09/09 - Siggi - PBS sgi feature directive only used if explicitly
	108	# set in the config file by the user
[399]	109	# 16/10/09 - Carolin - adjustments for archiving on SGI-ICE of binary files;
	110	# adding special1q
[1]	111
	112	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
[352]	113	delete_dayfile=false
	114	email_notification=none
[122]	115	group_number=none
[1]	116	locat=normal
	117	no_default_queue=none
	118	no_submit=false
	119	job_catalog="~/job_queue"
	120	job_name=none
	121	local_user=$LOGNAME
	122	node_usage=shared
	123	numprocs=1
	124	punkte="..........................................................."
	125	submcom=qsub
	126	queue=default
	127	remote_host=none
	128	remote_user=""
	129	verify=true
	130
	131	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
	132	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
	133	typeset -L20 spalte1
	134	typeset -R40 spalte2
	135	typeset -L60 spalte3
	136
	137
	138
	139	# FEHLERBEHANDLUNG
	140	# BEI EXIT:
	141	trap 'if [[ $locat != normal ]]
	142	then
	143	case $locat in
	144	(option) printf "\n --> available optios can be displayed"
	145	printf " by typing:"
	146	printf "\n \"subjob ?\" \n";;
	147	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	148	(*) printf "\n +++ unknown error"
	149	printf "\n please inform S. Raasch!\n"
	150	esac
	151	[[ -f $job_to_send ]] && rm $job_to_send
	152	printf "\n\n+++ SUBJOB killed \n\n"
	153	fi' exit
	154
	155
	156	# BEI TERMINAL-BREAK:
	157	trap '[[ -f $job_to_send ]] && rm $job_to_send
	158	printf "\n\n+++ SUBJOB killed \n\n"
	159	exit
	160	' 2
	161
	162
	163
	164
	165	# LOKALEN HOSTNAMEN FESTSTELLEN
	166	local_host=$(hostname)
	167
	168
	169
	170	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
	171	# UEBERHAUPT ZULAESSIG IST
	172	case $local_host in
	173	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[102]	174	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	175	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[210]	176	(bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
[291]	177	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[1]	178	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
[12]	179	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[292]	180	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
	181	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[372]	182	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[108]	183	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
[1]	184	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	185	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[108]	186	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
[367]	187	(node*) local_addres=165.132.26.61 local_host=lck;;
[101]	188	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[1]	189	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	190	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	191	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	192	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[164]	193	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
[292]	194	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
[210]	195	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
[312]	196	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[5]	197	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
	198	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
	199	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[164]	200	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
[1]	201	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	202	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	203	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	204	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[5]	205	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[1]	206	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
	207	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
[259]	208	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
[181]	209	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
[196]	210	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
[1]	211	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
	212	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
	213	(*) printf "\n +++ \"$local_host\" unknown";
	214	printf "\n please inform S. Raasch!";
	215	locat=parameter; exit;;
	216	esac
	217
	218
	219
	220	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	221	remote_host=$local_host
	222
	223
	224
	225
	226	# PROZEDUROPTIONEN EINLESEN
[352]	227	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
[1]	228	do
	229	case $option in
	230	(c) job_catalog=$OPTARG;;
	231	(d) delete_dayfile=true;;
	232	(D) no_submit=true;;
[352]	233	(e) email_notification=$OPTARG;;
[125]	234	(g) group_number=$OPTARG;;
[1]	235	(h) remote_host=$OPTARG;;
	236	(m) memory=$OPTARG;;
	237	(n) job_name=$OPTARG;;
	238	(N) node_usage=$OPTARG;;
	239	(O) threads_per_task=$OPTARG;;
	240	(q) no_default_queue=$OPTARG;;
	241	(t) cputime=$OPTARG;;
	242	(T) tasks_per_node=$OPTARG;;
	243	(u) remote_user=$OPTARG;;
	244	(v) verify=false;;
	245	(X) numprocs=$OPTARG;;
	246	(\?) printf "\n +++ Option $OPTARG unknown \n";
	247	locat=option; exit;;
	248	esac
	249	done
	250
	251
	252	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	253	shift OPTIND-1; file_to_send=$1
	254
	255
	256	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	257	if [ "$1" = "?" ]
	258	then
	259	(printf "\n *** subjob can be called as follows:\n"
	260	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	261	printf "\n Description of available options:\n"
	262	printf "\n Option Description Default-Value"
	263	printf "\n -c job-input- and output-catalog ~/job_queue"
	264	printf "\n -d no job-protocol will be created ---"
	265	printf "\n -D only the job-file will be created ---"
	266	printf "\n -h execution host, available hosts: $remote_host"
[5]	267	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
[251]	268	printf "\n lctit, nech, necriam, unics"
[1]	269	printf "\n -m memory demand per process in MByte ---"
	270	printf "\n -n jobname <jobdatei>"
	271	printf "\n -O threads per task (for OpenMP usage) 1"
	272	printf "\n -q job-queue to be used default"
	273	printf "\n -t allowed cpu-time in seconds ---"
	274	printf "\n -T tasks per node (on parallel hosts) ---"
	275	printf "\n -u username on execution host from .netrc"
	276	printf "\n -v no prompt for confirmation ---"
	277	printf "\n -X # of processors (on parallel hosts) 1"
	278	printf "\n "
	279	printf "\n The only possible positional parameter is <jobfile>:"
	280	printf "\n The complete NQS-job must be provided here."
	281	printf "\n <jobfile>=? creates this outline\n\n") \| more
	282	exit
	283	fi
	284
	285
	286
	287	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	288	if [[ "$file_to_send" = "" ]]
	289	then
	290	printf "\n +++ job-file missing"
	291	locat=parameter; exit
	292	else
	293	if [[ -f $file_to_send ]]
	294	then
	295	true
	296	else
	297	printf "\n +++ job-file: "
	298	printf "\n $file_to_send"
	299	printf "\n does not exist"
	300	locat=parameter; exit
	301	fi
	302	fi
	303
	304
	305
	306	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	307	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	308	if [[ $job_name = none ]]
	309	then
	310	job_name=$file_to_send
	311	fi
	312	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	313	then
	314	printf "\n +++ job-file name: "
	315	printf "\n $job_name"
	316	printf "\n must not contain \"/\"-characters"
	317	locat=parameter; exit
	318	fi
	319
	320
	321
	322
	323	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	324	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	325	if [[ $remote_host = none ]]
	326	then
	327	printf "\n +++ host missing"
	328	locat=option; exit
	329	else
	330	case $remote_host in
[251]	331	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	332	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[312]	333	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[251]	334	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	335	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	336	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
[291]	337	(lcsgib) queue=smallq; remote_addres=130.73.232.103; submcom=/opt/moab/bin/msub;;
[346]	338	(lcsgih) queue=smallq; remote_addres=130.75.4.102; submcom=/opt/moab/bin/msub;;
[367]	339	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[251]	340	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
	341	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
	342	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	343	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
	344	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	345	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	346	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	347	locat=parameter; exit;;
[1]	348	esac
	349	fi
	350
	351
	352	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	353	if [[ $no_default_queue != none ]]
	354	then
	355	error=false
	356	ndq=$no_default_queue
	357	case $remote_host in
	358	(ibm) case $ndq in
	359	(p690_express\|p690_standard\|p690_long) error=false;;
	360	(*) error=true;;
	361	esac;;
	362	(ibmb) case $ndq in
	363	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
	364	(*) error=true;;
	365	esac;;
	366	(ibmh) case $ndq in
[312]	367	(no_class) error=false;;
[1]	368	(*) error=true;;
	369	esac;;
	370	(ibms) case $ndq in
	371	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	372	(*) error=true;;
	373	esac;;
	374	(ibmy) case $ndq in
	375	(parallel) error=false;;
	376	(*) error=true;;
	377	esac;;
[191]	378	(lcsgib\|lcsgih) case $ndq in
[399]	379	(testq\|serialq\|smallq\|bigq\|workq\|dataq\|special1q) error=false;;
[191]	380	(*) error=true;;
	381	esac;;
[1]	382	(lctit) case $ndq in
[129]	383	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
[1]	384	(*) error=true;;
	385	esac;;
	386	(t3eb) case $ndq in
	387	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	388	(*) error=true;;
	389	esac;;
[251]	390	(necriam) case $ndq in
[253]	391	(SP\|SS\|P6) error=false;;
[251]	392	(*) error=true;;
	393	esac;;
[1]	394	(t3eh) case $ndq in
	395	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	396	(*) error=true;;
	397	esac;;
	398	(t3ej2\|t3ej5) case $ndq in
	399	(low\|normal\|high) error=false;;
	400	(*) error=true;;
	401	esac;;
	402	(t3es) case $ndq in
	403	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	404	(*) error=true;;
	405	esac;;
	406	(unics) case $ndq in
	407	(unics\|ht) error=false;;
	408	(*) error=true;;
	409	esac;;
	410	esac
	411	if [[ $error = true ]]
	412	then
	413	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	414	locat=parameter; exit
	415	else
	416	queue=$no_default_queue
	417	fi
	418	fi
	419
	420
	421
	422	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
	423	if [[ $node_usage != shared && $queue = cdev ]]
	424	then
	425	node_usage=shared
	426	fi
	427
	428
	429
	430	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	431	# AUFTEILEN
	432	done=false
	433	while [[ $done = false ]]
	434	do
	435	if (( $cputime <= 0 ))
	436	then
	437	printf "\n +++ wrong cpu-time or cpu-time missing"
	438	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	439	printf "\n >>> "
	440	read cputime 1>/dev/null 2>&1
	441	else
	442	done=true
	443	fi
	444	done
	445	if [[ $remote_host = nech ]]
	446	then
	447	if (( tasks_per_node != 0 ))
	448	then
	449	(( cputime = cputime * tasks_per_node ))
	450	elif [[ $numprocs != 0 ]]
	451	then
	452	(( cputime = cputime * numprocs ))
	453	fi
	454	fi
	455	(( stunden = cputime / 3600 ))
	456	(( resttime = cputime - stunden * 3600 ))
	457	(( minuten = resttime / 60 ))
	458	(( sekunden = resttime - minuten * 60 ))
	459	timestring=${stunden}:${minuten}:${sekunden}
	460
	461
	462
	463	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	464	done=false
	465	while [[ $done = false ]]
	466	do
	467	if (( memory <= 0 ))
	468	then
	469	printf "\n +++ wrong memory demand or memory demand missing"
	470	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	471	printf "\n >>> "
	472	read memory 1>/dev/null 2>&1
	473	else
	474	done=true
	475	fi
	476	done
	477
[251]	478	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
[1]	479	then
	480	if (( tasks_per_node != 0 ))
	481	then
	482	(( Memory = memory * tasks_per_node / 1000 ))
	483	elif [[ $numprocs != 0 ]]
	484	then
	485	(( Memory = memory * numprocs / 1000 ))
	486	else
	487	(( Memory = memory / 1000 ))
	488	fi
	489	fi
	490
	491
	492	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	493	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	494	then
	495	(( memory = memory * threads_per_task ))
	496	fi
	497
	498
	499	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	500	if (( tasks_per_node != 0 ))
	501	then
[377]	502	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	503	then
	504	(( nodes = numprocs / tasks_per_node ))
	505	else
	506	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
	507	fi
[1]	508	fi
	509
	510
	511
	512	# HEADER-AUSGABE
	513	if [[ $verify = true ]]
	514	then
	515	printf "\n\n"
	516	printf "#--------------------------------------------------------------# \n"
	517	spalte1=SUBJOB;spalte2=$(date)
	518	printf "\| $spalte1$spalte2 \| \n"
	519	printf "\| \| \n"
	520	printf "\| values of parameters/options: \| \n"
	521	spalte1=local_host$punkte; spalte2=$punkte$local_host
	522	printf "\| $spalte1$spalte2 \| \n"
	523	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	524	printf "\| $spalte1$spalte2 \| \n"
	525	spalte1=queue$punkte; spalte2=$punkte$queue
	526	printf "\| $spalte1$spalte2 \| \n"
	527	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	528	printf "\| $spalte1$spalte2 \| \n"
	529	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	530	printf "\| $spalte1$spalte2 \| \n"
	531	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	532	printf "\| $spalte1$spalte2 \| \n"
	533	printf "#--------------------------------------------------------------# \n\n"
	534
	535
	536	# KONTROLLABFRAGE, OB ALLES O.K.
	537	antwort="dummy"
	538	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	539	do
	540	read antwort?" >>> continue (y/n) ? "
	541	done
	542	if [[ $antwort = n \|\| $antwort = N ]]
	543	then
	544	locat=verify; exit
	545	fi
	546	printf "\n"
	547	fi
	548
	549
	550
	551	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	552	kennung=$RANDOM
	553	job_on_remhost=${job_name}_${kennung}_$local_host
	554	job_to_send=job_to_send_$kennung
	555	if [[ $delete_dayfile = false ]]
	556	then
	557	remote_dayfile=${local_host}_${job_name}_result_$kennung
	558	local_dayfile=${remote_host}_${job_name}
	559	else
	560	remote_dayfile=/dev/null
	561	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	562	then
	563	notification='# @ notification = never'
	564	fi
	565	fi
	566
	567
	568
	569	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
	570	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	571	then
	572
	573	if [[ $remote_host = ibmy ]]
	574	then
	575	consumable_memory=""
	576	else
	577	consumable_memory="ConsumableMemory($memory mb)"
	578	fi
[312]	579
	580	if [[ $remote_host = ibmh ]]
[1]	581	then
[312]	582	data_limit=""
	583	network_to_use=""
	584	class=""
	585	environment=""
[1]	586	else
[312]	587	class="# @ class = $queue"
	588	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	589	if [[ $queue = cdev ]]
[1]	590	then
[312]	591	data_limit="# @ data_limit = 1.76gb"
	592	network_to_use="# @ network.mpi = sn_all,shared,ip"
[1]	593	else
[312]	594	if [[ $remote_host = ibms ]]
	595	then
	596	network_to_use="# @ network.mpi = csss,shared,us"
	597	elif [[ $remote_host = ibmy ]]
	598	then
	599	network_to_use=""
	600	else
	601	network_to_use="# @ network.mpi = sn_all,shared,us"
	602	data_limit="# @ data_limit = 1.76gb"
	603	fi
[1]	604	fi
	605	fi
	606
	607	cat > $job_to_send << %%END%%
	608	#!/bin/ksh
[312]	609	# @ shell = /bin/ksh
[1]	610
	611	# @ job_type = parallel
	612	# @ job_name = palm
	613	# @ wall_clock_limit = ${timestring},$timestring
	614	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	615	# @ output = $remote_dayfile
	616	# @ error = $remote_dayfile
	617	# @ image_size = 50
[312]	618	$class
	619	$environment
[1]	620	$network_to_use
	621	$data_limit
	622	$notification
	623
	624	%%END%%
	625
	626	if (( nodes > 0 ))
	627	then
	628
	629	cat >> $job_to_send << %%END%%
	630	# @ node = $nodes
	631	# @ tasks_per_node = $tasks_per_node
	632	# @ node_usage = $node_usage
	633	# @ queue
	634
	635	%%END%%
	636
	637	else
	638
	639	if [[ $remote_host != ibmy ]]
	640	then
	641
	642	cat >> $job_to_send << %%END%%
	643	# @ blocking = unlimited
	644	# @ total_tasks = $numprocs
	645	# @ node_usage = $node_usage
	646	# @ queue
	647
	648	%%END%%
	649
	650	else
	651
	652	cat >> $job_to_send << %%END%%
	653	# @ node = 1
	654	# @ total_tasks = $numprocs
	655	# @ queue
	656
	657	%%END%%
	658
	659	fi
	660
	661	fi
	662
	663	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	664	then
	665
	666	cat > $job_to_send << %%END%%
	667	#!/bin/ksh
	668
	669	# @ job_type = serial
	670	# @ node_usage = $node_usage
	671	# @ job_name = palm
	672	# @ wall_clock_limit = ${timestring},$timestring
	673	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	674	# @ output = $remote_dayfile
	675	# @ error = $remote_dayfile
[312]	676	$class
[1]	677	$notification
	678
	679	# @ queue
	680
	681	%%END%%
	682
[108]	683	elif [[ $remote_host = lcfimm ]]
	684	then
	685
	686	if [[ $numprocs != 0 ]]
	687	then
	688	cat > $job_to_send << %%END%%
	689	#!/bin/ksh
	690	#PBS -N $job_name
[206]	691	#PBS -A geofysisk
[108]	692	#PBS -l walltime=$timestring
	693	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
	694	#PBS -l pmem=${memory}mb
	695	#PBS -m abe
[343]	696	#PBS -M maronga@muk.uni-hannover.de
[108]	697	#PBS -o $remote_dayfile
	698	#PBS -j oe
	699	mpd &
	700
	701	%%END%%
	702
[168]	703	else
	704	cat > $job_to_send << %%END%%
	705	#!/bin/ksh
	706	#PBS -N $job_name
[206]	707	#PBS -A geofysisk
[168]	708	#PBS -l walltime=$timestring
	709	#PBS -l ncpus=1
	710	#PBS -l pmem=${memory}mb
	711	#PBS -m abe
[343]	712	#PBS -M maronga@muk.uni-hannover.de
[168]	713	#PBS -o $remote_dayfile
	714	#PBS -j oe
	715
	716	%%END%%
	717
	718	fi
	719
[368]	720	elif [[ $remote_host = lck ]]
	721	then
	722
	723	if [[ $numprocs != 0 ]]
	724	then
	725	cat > $job_to_send << %%END%%
	726	#!/bin/ksh
	727	#PBS -N $job_name
	728	#PBS -l walltime=$timestring
	729	#PBS -l ncpus=$numprocs
	730	#PBS -l pmem=${memory}mb
	731	#PBS -o $remote_dayfile
	732	#PBS -j oe
	733
	734	mpd &
	735
	736	%%END%%
	737
	738	else
	739	cat > $job_to_send << %%END%%
	740	#!/bin/ksh
	741	#PBS -N $job_name
	742	#PBS -l walltime=$timestring
	743	#PBS -l ncpus=1
	744	#PBS -l pmem=${memory}mb
	745	#PBS -o $remote_dayfile
	746	#PBS -j oe
	747
	748	%%END%%
	749
	750	fi
	751
[181]	752	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
[179]	753	then
	754
[376]	755	if [[ "$sgi_feature" != "" ]]
	756	then
	757	feature_directive="#PBS -l feature=$sgi_feature"
	758	else
	759	feature_directive=""
	760	fi
[399]	761
	762	if [[ $queue = dataq ]]
	763	then
	764	feature_directive="#PBS -l feature=data"
	765	fi
[376]	766
[399]	767	if [[ $queue = testq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = special1q ]]
[191]	768	then
	769	queue_directive="#PBS -q $queue"
	770	else
	771	queue_directive=""
	772	fi
	773
[352]	774	if [[ $email_notification = none ]]
	775	then
	776	email_directive=""
	777	else
	778	email_directive="#PBS -M $email_notification"
	779	fi
	780
[179]	781	if [[ $numprocs != 0 ]]
	782	then
	783	cat > $job_to_send << %%END%%
[181]	784	#!/bin/bash
[179]	785	#PBS -N $job_name
	786	#PBS -l walltime=$timestring
[181]	787	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
	788	#PBS -l naccesspolicy=$node_usage
[179]	789	#PBS -o $remote_dayfile
	790	#PBS -j oe
[376]	791	$feature_directive
[191]	792	$queue_directive
[352]	793	$email_directive
[179]	794
	795	. /usr/share/modules/init/bash
[366]	796	module load ifort/11.0.069
[221]	797	module load $mpilib
[179]	798	module load netcdf
[374]	799
[181]	800	echo ld_library_path=\$LD_LIBRARY_PATH
[179]	801
	802	%%END%%
	803
	804	else
	805	cat > $job_to_send << %%END%%
	806	#PBS -S /bin/bash
	807	#PBS -N $job_name
	808	#PBS -l walltime=$timestring
	809	#PBS -l ncpus=1
	810	#PBS -o $remote_dayfile
	811	#PBS -e $remote_dayfile
	812
	813	%%END%%
	814
	815	fi
	816
[164]	817	elif [[ $remote_host = lcxt4 ]]
	818	then
	819
	820	if [[ $numprocs != 0 ]]
	821	then
	822	cat > $job_to_send << %%END%%
	823	#!/bin/ksh
	824	#PBS -S /bin/ksh
	825	#PBS -N $job_name
[206]	826	#PBS -A geofysisk
	827	#PBS -j oe
[164]	828	#PBS -l walltime=$timestring
	829	#PBS -l mppwidth=${numprocs}
	830	#PBS -l mppnppn=${tasks_per_node}
	831	#PBS -m abe
[343]	832	#PBS -M maronga@muk.uni-hannover.de
[164]	833	#PBS -o $remote_dayfile
	834
[343]	835	. /opt/modules/default/init/ksh
	836	module load modules/3.1.6
	837	module load PrgEnv-pgi/2.1.56HD
	838	module load Base-opts/2.1.56HD
	839	module load moab/5.2.5-s13441
	840	module load torque/2.3.4-snap.200809221601
	841
[164]	842	%%END%%
	843
[108]	844	else
	845	cat > $job_to_send << %%END%%
	846	#!/bin/ksh
[168]	847	#PBS -S /bin/ksh
[108]	848	#PBS -N $job_name
[206]	849	#PBS -A geofysisk
	850	#PBS -j oe
[108]	851	#PBS -l walltime=$timestring
	852	#PBS -l ncpus=1
	853	#PBS -l pmem=${memory}mb
	854	#PBS -m abe
[343]	855	#PBS -M maronga@muk.uni-hannover.de
[108]	856	#PBS -o $remote_dayfile
	857
[343]	858	. /opt/modules/default/init/ksh
	859	module load modules/3.1.6
	860	module load PrgEnv-pgi/2.1.56HD
	861	module load Base-opts/2.1.56HD
	862	module load moab/5.2.5-s13441
	863	module load torque/2.3.4-snap.200809221601
	864
[108]	865	%%END%%
	866
	867	fi
	868
[1]	869	elif [[ $remote_host = nech ]]
	870	then
	871
	872	if (( nodes > 1 ))
	873	then
	874	cat > $job_to_send << %%END%%
	875	#!/bin/ksh
	876	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	877	#PBS -l ${qsubmem}=${Memory}gb
	878	#PBS -b $nodes
	879	#PBS -o $remote_dayfile
	880	#PBS -N palm
	881	#PBS -j o
	882	#PBS -T mpisx
	883
	884	%%END%%
	885
	886	elif [[ $numprocs != 0 ]]
	887	then
	888	cat > $job_to_send << %%END%%
	889	#!/bin/ksh
	890	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	891	#PBS -l ${qsubmem}=${Memory}gb
	892	#PBS -o $remote_dayfile
	893	#PBS -N palm
	894	#PBS -j o
	895
	896	%%END%%
	897
	898	else
	899	cat > $job_to_send << %%END%%
	900	#!/bin/ksh
	901	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	902	#PBS -o $remote_dayfile
	903	#PBS -j o
	904
	905	%%END%%
	906
	907	fi
	908
[251]	909	elif [[ $remote_host = necriam ]]
[35]	910	then
	911
	912	if (( nodes > 1 ))
	913	then
	914	cat > $job_to_send << %%END%%
	915	#!/bin/ksh
	916	#PBS -b $nodes
	917	#PBS -o $remote_dayfile
[253]	918	#PBS -N $job_name
[35]	919	#PBS -j o
[251]	920	#PBS -v MPIPROGINV=YES
[35]	921
	922	%%END%%
	923
	924	elif [[ $numprocs != 0 ]]
	925	then
	926	cat > $job_to_send << %%END%%
	927	#!/bin/ksh
	928	#PBS -q ${queue}
	929	#PBS -o $remote_dayfile
[253]	930	#PBS -N $job_name
[35]	931	#PBS -j o
[251]	932	#PBS -v MPIPROGINV=YES
[35]	933
	934	%%END%%
	935
	936	else
	937	cat > $job_to_send << %%END%%
	938	#!/bin/ksh
	939	#PBS -o $remote_dayfile
	940	#PBS -j o
	941
	942	%%END%%
	943
	944	fi
	945
[1]	946	elif [[ $remote_host = lctit ]]
	947	then
	948	cat > $job_to_send << %%END%%
	949	#!/bin/bash
	950	#$ -S /bin/bash
	951	cd $job_catalog
[69]	952	export PATH=$PALM_BIN:\$PATH
	953	echo \$PATH
[1]	954
	955	%%END%%
	956
	957	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
[118]	958	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
[122]	959	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
[1]	960
	961	else
	962
	963	cat > $job_to_send << %%END%%
	964	# @\$-q ${queue}
	965	# @\$-l${qsubtime} $timestring
	966	# @\$-l${qsubmem} ${memory}mb
	967	# @\$-o $remote_dayfile
	968	# @\$-eo
	969
	970	%%END%%
	971
	972	fi
	973
	974
	975	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	976	# BEI EXIT VERANLASSEN
	977	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	978	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	979	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	980	then
	981	echo "set +vx" >> $job_to_send
	982	echo "trap '" >> $job_to_send
	983	echo "set +vx" >> $job_to_send
[181]	984	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]	985	then
	986	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
	987	then
	988	return_queue=c1
	989	elif [[ $remote_host = ibms ]]
	990	then
	991	return_queue=p_normal
	992	elif [[ $remote_host = ibmy ]]
	993	then
	994	return_queue=serial
[325]	995	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
	996	then
[326]	997	return_queue=serialq
[1]	998	elif [[ $remote_host = lctit ]]
	999	then
[118]	1000	return_queue=sla3
[251]	1001	elif [[ $remote_host = necriam ]]
[1]	1002	then
[253]	1003	return_queue=SP
[1]	1004	else
	1005	return_queue=unknown
	1006	fi
	1007
	1008	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1009	then
	1010
	1011	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1012	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	1013	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	1014	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	1015	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	1016	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	1017	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
[312]	1018	if [[ $host != "ibmh" ]]
	1019	then
	1020	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	1021	fi
[1]	1022	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	1023	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	1024
	1025	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	1026	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	1027
	1028	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	1029	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
	1030	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	1031
	1032	elif [[ $remote_host = nech ]]
	1033	then
	1034	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1035	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1036	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1037	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1038	echo "#PBS -j o" >> $job_to_send
	1039	echo " " >> $job_to_send
	1040	echo "set -x" >> $job_to_send
	1041	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1042	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1043	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1044	echo "%%END%%" >> $job_to_send
	1045
[251]	1046	elif [[ $remote_host = necriam ]]
[35]	1047	then
	1048	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1049	echo "#PBS -q $return_queue" >> $job_to_send
	1050	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1051	echo "#PBS -j o" >> $job_to_send
	1052	echo " " >> $job_to_send
	1053	echo "set -x" >> $job_to_send
	1054	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1055	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1056	echo "%%END%%" >> $job_to_send
	1057
[181]	1058	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1059	then
	1060	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1061	echo "#!/bin/bash" >> $job_to_send
	1062	echo "#PBS -N job_protocol_transfer" >> $job_to_send
[325]	1063	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[181]	1064	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[366]	1065	echo "#PBS -l feature=data" >> $job_to_send
[181]	1066	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1067	echo "#PBS -j oe" >> $job_to_send
	1068	echo " " >> $job_to_send
	1069	echo ". /usr/share/modules/init/bash" >> $job_to_send
	1070	echo "set -x" >> $job_to_send
	1071	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1072	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1073	echo "%%END%%" >> $job_to_send
	1074
[1]	1075	elif [[ $remote_host = lctit ]]
	1076	then
	1077	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1078	echo "set -x" >> $job_to_send
	1079	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
	1080	echo "cd $job_catalog" >> $job_to_send
	1081	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1082	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1083	echo "%%END%%" >> $job_to_send
	1084
	1085	else
	1086
	1087	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1088	echo "# @\\\$-q $return_queue" >> $job_to_send
	1089	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1090	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1091	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1092	then
	1093	echo "# @\$-l mpp_p=0" >> $job_to_send
	1094	fi
	1095	echo '# @\$-lF 10mb' >> $job_to_send
	1096	# echo '# @\$-o /dev/null' >> $job_to_send
	1097	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1098	echo '# @\\\$-eo' >> $job_to_send
	1099	echo " " >> $job_to_send
	1100	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1101	then
	1102	echo "set +vx" >> $job_to_send
	1103	echo ". .profile" >> $job_to_send
	1104	fi
	1105	echo "set -x" >> $job_to_send
	1106	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	1107	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1108	echo "%%END%%" >> $job_to_send
	1109	fi
	1110
	1111	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1112	then
	1113	echo "llsubmit scpjob.$kennung" >> $job_to_send
[181]	1114	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1115	then
	1116	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1117	echo "msub scpjob.$kennung" >> $job_to_send
[1]	1118	elif [[ $remote_host = lctit ]]
	1119	then
	1120	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1121	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
	1122	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
	1123	echo "sleep 10" >> $job_to_send
	1124	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1125	then
	1126	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	1127	elif [[ $remote_host = t3es ]]
	1128	then
	1129	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
	1130	else
	1131	echo "qsub scpjob.$kennung" >> $job_to_send
	1132	fi
	1133	echo "rm scpjob.$kennung" >> $job_to_send
	1134	if [[ $remote_host = nech ]]
	1135	then
	1136	echo "cd -" >> $job_to_send
	1137	fi
	1138	else
	1139	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1140	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1141	fi
	1142	echo "set -x" >> $job_to_send
	1143	echo " ' exit" >> $job_to_send
	1144	echo "set -x" >> $job_to_send
	1145	fi
	1146
	1147
	1148
	1149	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	1150	cat $file_to_send >> $job_to_send
[69]	1151
[1]	1152	if [[ $remote_host = ibm ]]
	1153	then
	1154	echo " " >> $job_to_send
	1155	echo "exit" >> $job_to_send
	1156	fi
[69]	1157	if [[ $remote_host = lctit ]]
	1158	then
	1159	echo " " >> $job_to_send
	1160	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1161	fi
[1]	1162
	1163
	1164
	1165	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	1166	if [[ -z $remote_user ]]
	1167	then
	1168	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	1169	then
	1170	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	1171	fi
	1172	fi
	1173
	1174
	1175
	1176	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	1177	if [[ $no_submit = false ]]
	1178	then
	1179	if [[ $remote_host != $local_host ]]
	1180	then
	1181	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	1182	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	1183	then
[82]	1184	job_catalog_save=$job_catalog
	1185	job_catalog=job_queue
	1186	elif [[ $remote_host = nech ]]
	1187	then
	1188	job_catalog_save=$job_catalog
	1189	job_catalog=/hpf/b/${remote_user}/job_queue
	1190	fi
	1191	if [[ $local_host = decalpha ]]
	1192	then
	1193	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	1194	# PASSWORT AB
	1195	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	1196	elif [[ $remote_host = nech ]]
	1197	then
	1198	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	1199	# TRANSFERIERT WERDEN
	1200	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1201	else
[82]	1202	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1203	fi
[82]	1204	if [[ $? = 1 ]]
	1205	then
	1206	locat=scp; exit
	1207	fi
	1208	if [[ $remote_host = ibms ]]
	1209	then
	1210	job_catalog=$job_catalog_save
	1211	fi
[1]	1212	[[ $verify = true ]] && printf "\n >>> finished\n"
	1213	else
	1214	eval job_catalog=$job_catalog
	1215	cp $job_to_send ${job_catalog}/$job_on_remhost
	1216	fi
	1217
	1218
	1219
	1220	# NQS- BZW. LOADLEVELER-JOB STARTEN
	1221	if [[ $remote_host != $local_host ]]
	1222	then
	1223	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[82]	1224	if [[ $remote_host != lctit ]]
[1]	1225	then
[372]	1226	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
	1227	then
	1228	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
	1229	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
	1230	else
	1231	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
	1232	fi
[1]	1233	else
[206]	1234	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
[82]	1235	# MIT SSH, DESHALB AUFRUF PER PIPE
	1236	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
	1237	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	1238	echo "first try" > send_protocol
	1239	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
	1240	do
	1241	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
	1242	then
	1243	printf "\n trying to submit job to TIT cluster..."
	1244	else
	1245	printf "\n +++ failed ... trying to submit again ..."
	1246	fi
	1247	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
	1248	cat send_protocol
	1249	done
	1250	sleep 10
	1251	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	1252	echo "$submcom $job_on_remhost"
	1253	rm send_protocol
	1254	fi
[1]	1255
	1256	[[ $verify = true ]] && printf " >>> o.k.\n"
	1257	else
	1258	cd $job_catalog
[210]	1259	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
[1]	1260	then
	1261	eval $submcom $job_on_remhost
[367]	1262	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 \|\| $localhost = lck ]]
[108]	1263	then
	1264	eval $submcom $job_on_remhost
	1265	echo "$submcom $job_on_remhost"
	1266	chmod u+x $job_on_remhost
[292]	1267	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
	1268	then
	1269	eval $submcom $job_on_remhost
[1]	1270	elif [[ $local_host = nech ]]
	1271	then
	1272	if [[ $queue = default ]]
	1273	then
	1274	eval $submcom $job_on_remhost
	1275	else
	1276	eval $submcom -q $queue $job_on_remhost
	1277	fi
	1278	else
	1279	qsub $job_on_remhost
	1280	fi
[69]	1281	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
	1282	[[ $local_host != lctit ]] && rm $job_on_remhost
[1]	1283	cd - > /dev/null
	1284	fi
	1285	fi
	1286
	1287
	1288
	1289	# ABSCHLUSSARBEITEN
	1290
	1291	if [[ $no_submit = false ]]
	1292	then
	1293	rm $job_to_send
	1294	fi
	1295	[[ $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 |