Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 402

Last change on this file since 402 was 400, checked in by letzel, 15 years ago
Add lcsgih\|b dataservers as local hosts, queue adjustments (subjob)
Property svn:keywords set to `Id Rev`
File size: 45.2 KB

Rev	Line
[1]	1	#!/bin/ksh
[82]	2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
[169]	3	# $Id: subjob 400 2009-10-20 13:34:43Z maronga $
[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;;
[400]	176	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
	177	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
	178	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
	179	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
	180	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
[291]	181	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[1]	182	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
[12]	183	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[292]	184	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[372]	185	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[108]	186	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
[1]	187	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	188	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[108]	189	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
[367]	190	(node*) local_addres=165.132.26.61 local_host=lck;;
[101]	191	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[1]	192	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	193	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	194	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	195	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[164]	196	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
[400]	197	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
	198	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
	199	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
	200	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
[292]	201	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
[210]	202	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
[312]	203	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[5]	204	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
	205	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
	206	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[164]	207	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
[1]	208	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	209	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	210	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	211	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[5]	212	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[1]	213	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
	214	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
[259]	215	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
[181]	216	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
[196]	217	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
[1]	218	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
	219	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
	220	(*) printf "\n +++ \"$local_host\" unknown";
	221	printf "\n please inform S. Raasch!";
	222	locat=parameter; exit;;
	223	esac
	224
	225
	226
	227	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	228	remote_host=$local_host
	229
	230
	231
	232
	233	# PROZEDUROPTIONEN EINLESEN
[352]	234	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
[1]	235	do
	236	case $option in
	237	(c) job_catalog=$OPTARG;;
	238	(d) delete_dayfile=true;;
	239	(D) no_submit=true;;
[352]	240	(e) email_notification=$OPTARG;;
[125]	241	(g) group_number=$OPTARG;;
[1]	242	(h) remote_host=$OPTARG;;
	243	(m) memory=$OPTARG;;
	244	(n) job_name=$OPTARG;;
	245	(N) node_usage=$OPTARG;;
	246	(O) threads_per_task=$OPTARG;;
	247	(q) no_default_queue=$OPTARG;;
	248	(t) cputime=$OPTARG;;
	249	(T) tasks_per_node=$OPTARG;;
	250	(u) remote_user=$OPTARG;;
	251	(v) verify=false;;
	252	(X) numprocs=$OPTARG;;
	253	(\?) printf "\n +++ Option $OPTARG unknown \n";
	254	locat=option; exit;;
	255	esac
	256	done
	257
	258
	259	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	260	shift OPTIND-1; file_to_send=$1
	261
	262
	263	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	264	if [ "$1" = "?" ]
	265	then
	266	(printf "\n *** subjob can be called as follows:\n"
	267	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	268	printf "\n Description of available options:\n"
	269	printf "\n Option Description Default-Value"
	270	printf "\n -c job-input- and output-catalog ~/job_queue"
	271	printf "\n -d no job-protocol will be created ---"
	272	printf "\n -D only the job-file will be created ---"
	273	printf "\n -h execution host, available hosts: $remote_host"
[5]	274	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
[251]	275	printf "\n lctit, nech, necriam, unics"
[1]	276	printf "\n -m memory demand per process in MByte ---"
	277	printf "\n -n jobname <jobdatei>"
	278	printf "\n -O threads per task (for OpenMP usage) 1"
	279	printf "\n -q job-queue to be used default"
	280	printf "\n -t allowed cpu-time in seconds ---"
	281	printf "\n -T tasks per node (on parallel hosts) ---"
	282	printf "\n -u username on execution host from .netrc"
	283	printf "\n -v no prompt for confirmation ---"
	284	printf "\n -X # of processors (on parallel hosts) 1"
	285	printf "\n "
	286	printf "\n The only possible positional parameter is <jobfile>:"
	287	printf "\n The complete NQS-job must be provided here."
	288	printf "\n <jobfile>=? creates this outline\n\n") \| more
	289	exit
	290	fi
	291
	292
	293
	294	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	295	if [[ "$file_to_send" = "" ]]
	296	then
	297	printf "\n +++ job-file missing"
	298	locat=parameter; exit
	299	else
	300	if [[ -f $file_to_send ]]
	301	then
	302	true
	303	else
	304	printf "\n +++ job-file: "
	305	printf "\n $file_to_send"
	306	printf "\n does not exist"
	307	locat=parameter; exit
	308	fi
	309	fi
	310
	311
	312
	313	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	314	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	315	if [[ $job_name = none ]]
	316	then
	317	job_name=$file_to_send
	318	fi
	319	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	320	then
	321	printf "\n +++ job-file name: "
	322	printf "\n $job_name"
	323	printf "\n must not contain \"/\"-characters"
	324	locat=parameter; exit
	325	fi
	326
	327
	328
	329
	330	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	331	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	332	if [[ $remote_host = none ]]
	333	then
	334	printf "\n +++ host missing"
	335	locat=option; exit
	336	else
	337	case $remote_host in
[251]	338	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	339	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[312]	340	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[251]	341	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	342	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	343	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
[291]	344	(lcsgib) queue=smallq; remote_addres=130.73.232.103; submcom=/opt/moab/bin/msub;;
[346]	345	(lcsgih) queue=smallq; remote_addres=130.75.4.102; submcom=/opt/moab/bin/msub;;
[367]	346	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[251]	347	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
	348	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
	349	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	350	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
	351	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	352	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	353	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	354	locat=parameter; exit;;
[1]	355	esac
	356	fi
	357
	358
	359	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	360	if [[ $no_default_queue != none ]]
	361	then
	362	error=false
	363	ndq=$no_default_queue
	364	case $remote_host in
	365	(ibm) case $ndq in
	366	(p690_express\|p690_standard\|p690_long) error=false;;
	367	(*) error=true;;
	368	esac;;
	369	(ibmb) case $ndq in
	370	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
	371	(*) error=true;;
	372	esac;;
	373	(ibmh) case $ndq in
[312]	374	(no_class) error=false;;
[1]	375	(*) error=true;;
	376	esac;;
	377	(ibms) case $ndq in
	378	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	379	(*) error=true;;
	380	esac;;
	381	(ibmy) case $ndq in
	382	(parallel) error=false;;
	383	(*) error=true;;
	384	esac;;
[191]	385	(lcsgib\|lcsgih) case $ndq in
[400]	386	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|special1q) error=false;;
[191]	387	(*) error=true;;
	388	esac;;
[1]	389	(lctit) case $ndq in
[129]	390	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
[1]	391	(*) error=true;;
	392	esac;;
	393	(t3eb) case $ndq in
	394	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	395	(*) error=true;;
	396	esac;;
[251]	397	(necriam) case $ndq in
[253]	398	(SP\|SS\|P6) error=false;;
[251]	399	(*) error=true;;
	400	esac;;
[1]	401	(t3eh) case $ndq in
	402	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	403	(*) error=true;;
	404	esac;;
	405	(t3ej2\|t3ej5) case $ndq in
	406	(low\|normal\|high) error=false;;
	407	(*) error=true;;
	408	esac;;
	409	(t3es) case $ndq in
	410	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	411	(*) error=true;;
	412	esac;;
	413	(unics) case $ndq in
	414	(unics\|ht) error=false;;
	415	(*) error=true;;
	416	esac;;
	417	esac
	418	if [[ $error = true ]]
	419	then
	420	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	421	locat=parameter; exit
	422	else
	423	queue=$no_default_queue
	424	fi
	425	fi
	426
	427
	428
	429	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
	430	if [[ $node_usage != shared && $queue = cdev ]]
	431	then
	432	node_usage=shared
	433	fi
	434
	435
	436
	437	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	438	# AUFTEILEN
	439	done=false
	440	while [[ $done = false ]]
	441	do
	442	if (( $cputime <= 0 ))
	443	then
	444	printf "\n +++ wrong cpu-time or cpu-time missing"
	445	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	446	printf "\n >>> "
	447	read cputime 1>/dev/null 2>&1
	448	else
	449	done=true
	450	fi
	451	done
	452	if [[ $remote_host = nech ]]
	453	then
	454	if (( tasks_per_node != 0 ))
	455	then
	456	(( cputime = cputime * tasks_per_node ))
	457	elif [[ $numprocs != 0 ]]
	458	then
	459	(( cputime = cputime * numprocs ))
	460	fi
	461	fi
	462	(( stunden = cputime / 3600 ))
	463	(( resttime = cputime - stunden * 3600 ))
	464	(( minuten = resttime / 60 ))
	465	(( sekunden = resttime - minuten * 60 ))
	466	timestring=${stunden}:${minuten}:${sekunden}
	467
	468
	469
	470	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	471	done=false
	472	while [[ $done = false ]]
	473	do
	474	if (( memory <= 0 ))
	475	then
	476	printf "\n +++ wrong memory demand or memory demand missing"
	477	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	478	printf "\n >>> "
	479	read memory 1>/dev/null 2>&1
	480	else
	481	done=true
	482	fi
	483	done
	484
[251]	485	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
[1]	486	then
	487	if (( tasks_per_node != 0 ))
	488	then
	489	(( Memory = memory * tasks_per_node / 1000 ))
	490	elif [[ $numprocs != 0 ]]
	491	then
	492	(( Memory = memory * numprocs / 1000 ))
	493	else
	494	(( Memory = memory / 1000 ))
	495	fi
	496	fi
	497
	498
	499	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	500	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	501	then
	502	(( memory = memory * threads_per_task ))
	503	fi
	504
	505
	506	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	507	if (( tasks_per_node != 0 ))
	508	then
[377]	509	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	510	then
	511	(( nodes = numprocs / tasks_per_node ))
	512	else
	513	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
	514	fi
[1]	515	fi
	516
	517
	518
	519	# HEADER-AUSGABE
	520	if [[ $verify = true ]]
	521	then
	522	printf "\n\n"
	523	printf "#--------------------------------------------------------------# \n"
	524	spalte1=SUBJOB;spalte2=$(date)
	525	printf "\| $spalte1$spalte2 \| \n"
	526	printf "\| \| \n"
	527	printf "\| values of parameters/options: \| \n"
	528	spalte1=local_host$punkte; spalte2=$punkte$local_host
	529	printf "\| $spalte1$spalte2 \| \n"
	530	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	531	printf "\| $spalte1$spalte2 \| \n"
	532	spalte1=queue$punkte; spalte2=$punkte$queue
	533	printf "\| $spalte1$spalte2 \| \n"
	534	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	535	printf "\| $spalte1$spalte2 \| \n"
	536	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	537	printf "\| $spalte1$spalte2 \| \n"
	538	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	539	printf "\| $spalte1$spalte2 \| \n"
	540	printf "#--------------------------------------------------------------# \n\n"
	541
	542
	543	# KONTROLLABFRAGE, OB ALLES O.K.
	544	antwort="dummy"
	545	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	546	do
	547	read antwort?" >>> continue (y/n) ? "
	548	done
	549	if [[ $antwort = n \|\| $antwort = N ]]
	550	then
	551	locat=verify; exit
	552	fi
	553	printf "\n"
	554	fi
	555
	556
	557
	558	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	559	kennung=$RANDOM
	560	job_on_remhost=${job_name}_${kennung}_$local_host
	561	job_to_send=job_to_send_$kennung
	562	if [[ $delete_dayfile = false ]]
	563	then
	564	remote_dayfile=${local_host}_${job_name}_result_$kennung
	565	local_dayfile=${remote_host}_${job_name}
	566	else
	567	remote_dayfile=/dev/null
	568	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	569	then
	570	notification='# @ notification = never'
	571	fi
	572	fi
	573
	574
	575
	576	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
	577	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	578	then
	579
	580	if [[ $remote_host = ibmy ]]
	581	then
	582	consumable_memory=""
	583	else
	584	consumable_memory="ConsumableMemory($memory mb)"
	585	fi
[312]	586
	587	if [[ $remote_host = ibmh ]]
[1]	588	then
[312]	589	data_limit=""
	590	network_to_use=""
	591	class=""
	592	environment=""
[1]	593	else
[312]	594	class="# @ class = $queue"
	595	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	596	if [[ $queue = cdev ]]
[1]	597	then
[312]	598	data_limit="# @ data_limit = 1.76gb"
	599	network_to_use="# @ network.mpi = sn_all,shared,ip"
[1]	600	else
[312]	601	if [[ $remote_host = ibms ]]
	602	then
	603	network_to_use="# @ network.mpi = csss,shared,us"
	604	elif [[ $remote_host = ibmy ]]
	605	then
	606	network_to_use=""
	607	else
	608	network_to_use="# @ network.mpi = sn_all,shared,us"
	609	data_limit="# @ data_limit = 1.76gb"
	610	fi
[1]	611	fi
	612	fi
	613
	614	cat > $job_to_send << %%END%%
	615	#!/bin/ksh
[312]	616	# @ shell = /bin/ksh
[1]	617
	618	# @ job_type = parallel
	619	# @ job_name = palm
	620	# @ wall_clock_limit = ${timestring},$timestring
	621	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	622	# @ output = $remote_dayfile
	623	# @ error = $remote_dayfile
	624	# @ image_size = 50
[312]	625	$class
	626	$environment
[1]	627	$network_to_use
	628	$data_limit
	629	$notification
	630
	631	%%END%%
	632
	633	if (( nodes > 0 ))
	634	then
	635
	636	cat >> $job_to_send << %%END%%
	637	# @ node = $nodes
	638	# @ tasks_per_node = $tasks_per_node
	639	# @ node_usage = $node_usage
	640	# @ queue
	641
	642	%%END%%
	643
	644	else
	645
	646	if [[ $remote_host != ibmy ]]
	647	then
	648
	649	cat >> $job_to_send << %%END%%
	650	# @ blocking = unlimited
	651	# @ total_tasks = $numprocs
	652	# @ node_usage = $node_usage
	653	# @ queue
	654
	655	%%END%%
	656
	657	else
	658
	659	cat >> $job_to_send << %%END%%
	660	# @ node = 1
	661	# @ total_tasks = $numprocs
	662	# @ queue
	663
	664	%%END%%
	665
	666	fi
	667
	668	fi
	669
	670	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	671	then
	672
	673	cat > $job_to_send << %%END%%
	674	#!/bin/ksh
	675
	676	# @ job_type = serial
	677	# @ node_usage = $node_usage
	678	# @ job_name = palm
	679	# @ wall_clock_limit = ${timestring},$timestring
	680	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	681	# @ output = $remote_dayfile
	682	# @ error = $remote_dayfile
[312]	683	$class
[1]	684	$notification
	685
	686	# @ queue
	687
	688	%%END%%
	689
[108]	690	elif [[ $remote_host = lcfimm ]]
	691	then
	692
	693	if [[ $numprocs != 0 ]]
	694	then
	695	cat > $job_to_send << %%END%%
	696	#!/bin/ksh
	697	#PBS -N $job_name
[206]	698	#PBS -A geofysisk
[108]	699	#PBS -l walltime=$timestring
	700	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
	701	#PBS -l pmem=${memory}mb
	702	#PBS -m abe
[343]	703	#PBS -M maronga@muk.uni-hannover.de
[108]	704	#PBS -o $remote_dayfile
	705	#PBS -j oe
	706	mpd &
	707
	708	%%END%%
	709
[168]	710	else
	711	cat > $job_to_send << %%END%%
	712	#!/bin/ksh
	713	#PBS -N $job_name
[206]	714	#PBS -A geofysisk
[168]	715	#PBS -l walltime=$timestring
	716	#PBS -l ncpus=1
	717	#PBS -l pmem=${memory}mb
	718	#PBS -m abe
[343]	719	#PBS -M maronga@muk.uni-hannover.de
[168]	720	#PBS -o $remote_dayfile
	721	#PBS -j oe
	722
	723	%%END%%
	724
	725	fi
	726
[368]	727	elif [[ $remote_host = lck ]]
	728	then
	729
	730	if [[ $numprocs != 0 ]]
	731	then
	732	cat > $job_to_send << %%END%%
	733	#!/bin/ksh
	734	#PBS -N $job_name
	735	#PBS -l walltime=$timestring
	736	#PBS -l ncpus=$numprocs
	737	#PBS -l pmem=${memory}mb
	738	#PBS -o $remote_dayfile
	739	#PBS -j oe
	740
	741	mpd &
	742
	743	%%END%%
	744
	745	else
	746	cat > $job_to_send << %%END%%
	747	#!/bin/ksh
	748	#PBS -N $job_name
	749	#PBS -l walltime=$timestring
	750	#PBS -l ncpus=1
	751	#PBS -l pmem=${memory}mb
	752	#PBS -o $remote_dayfile
	753	#PBS -j oe
	754
	755	%%END%%
	756
	757	fi
	758
[181]	759	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
[179]	760	then
	761
[376]	762	if [[ "$sgi_feature" != "" ]]
	763	then
	764	feature_directive="#PBS -l feature=$sgi_feature"
	765	else
	766	feature_directive=""
	767	fi
[399]	768
	769	if [[ $queue = dataq ]]
	770	then
	771	feature_directive="#PBS -l feature=data"
	772	fi
[376]	773
[400]	774	if [[ $queue = testq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = serialq \|\| $queue = special1q ]]
[191]	775	then
	776	queue_directive="#PBS -q $queue"
	777	else
	778	queue_directive=""
	779	fi
	780
[352]	781	if [[ $email_notification = none ]]
	782	then
	783	email_directive=""
	784	else
	785	email_directive="#PBS -M $email_notification"
	786	fi
	787
[179]	788	if [[ $numprocs != 0 ]]
	789	then
	790	cat > $job_to_send << %%END%%
[181]	791	#!/bin/bash
[179]	792	#PBS -N $job_name
	793	#PBS -l walltime=$timestring
[181]	794	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
	795	#PBS -l naccesspolicy=$node_usage
[179]	796	#PBS -o $remote_dayfile
	797	#PBS -j oe
[376]	798	$feature_directive
[191]	799	$queue_directive
[352]	800	$email_directive
[179]	801
	802	. /usr/share/modules/init/bash
[366]	803	module load ifort/11.0.069
[221]	804	module load $mpilib
[179]	805	module load netcdf
[374]	806
[181]	807	echo ld_library_path=\$LD_LIBRARY_PATH
[179]	808
	809	%%END%%
	810
	811	else
	812	cat > $job_to_send << %%END%%
	813	#PBS -S /bin/bash
	814	#PBS -N $job_name
	815	#PBS -l walltime=$timestring
	816	#PBS -l ncpus=1
	817	#PBS -o $remote_dayfile
	818	#PBS -e $remote_dayfile
[400]	819	$queue_directive
	820	$email_directive
[179]	821
	822	%%END%%
	823
	824	fi
	825
[164]	826	elif [[ $remote_host = lcxt4 ]]
	827	then
	828
	829	if [[ $numprocs != 0 ]]
	830	then
	831	cat > $job_to_send << %%END%%
	832	#!/bin/ksh
	833	#PBS -S /bin/ksh
	834	#PBS -N $job_name
[206]	835	#PBS -A geofysisk
	836	#PBS -j oe
[164]	837	#PBS -l walltime=$timestring
	838	#PBS -l mppwidth=${numprocs}
	839	#PBS -l mppnppn=${tasks_per_node}
	840	#PBS -m abe
[343]	841	#PBS -M maronga@muk.uni-hannover.de
[164]	842	#PBS -o $remote_dayfile
	843
[343]	844	. /opt/modules/default/init/ksh
	845	module load modules/3.1.6
	846	module load PrgEnv-pgi/2.1.56HD
	847	module load Base-opts/2.1.56HD
	848	module load moab/5.2.5-s13441
	849	module load torque/2.3.4-snap.200809221601
	850
[164]	851	%%END%%
	852
[108]	853	else
	854	cat > $job_to_send << %%END%%
	855	#!/bin/ksh
[168]	856	#PBS -S /bin/ksh
[108]	857	#PBS -N $job_name
[206]	858	#PBS -A geofysisk
	859	#PBS -j oe
[108]	860	#PBS -l walltime=$timestring
	861	#PBS -l ncpus=1
	862	#PBS -l pmem=${memory}mb
	863	#PBS -m abe
[343]	864	#PBS -M maronga@muk.uni-hannover.de
[108]	865	#PBS -o $remote_dayfile
	866
[343]	867	. /opt/modules/default/init/ksh
	868	module load modules/3.1.6
	869	module load PrgEnv-pgi/2.1.56HD
	870	module load Base-opts/2.1.56HD
	871	module load moab/5.2.5-s13441
	872	module load torque/2.3.4-snap.200809221601
	873
[108]	874	%%END%%
	875
	876	fi
	877
[1]	878	elif [[ $remote_host = nech ]]
	879	then
	880
	881	if (( nodes > 1 ))
	882	then
	883	cat > $job_to_send << %%END%%
	884	#!/bin/ksh
	885	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	886	#PBS -l ${qsubmem}=${Memory}gb
	887	#PBS -b $nodes
	888	#PBS -o $remote_dayfile
	889	#PBS -N palm
	890	#PBS -j o
	891	#PBS -T mpisx
	892
	893	%%END%%
	894
	895	elif [[ $numprocs != 0 ]]
	896	then
	897	cat > $job_to_send << %%END%%
	898	#!/bin/ksh
	899	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	900	#PBS -l ${qsubmem}=${Memory}gb
	901	#PBS -o $remote_dayfile
	902	#PBS -N palm
	903	#PBS -j o
	904
	905	%%END%%
	906
	907	else
	908	cat > $job_to_send << %%END%%
	909	#!/bin/ksh
	910	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	911	#PBS -o $remote_dayfile
	912	#PBS -j o
	913
	914	%%END%%
	915
	916	fi
	917
[251]	918	elif [[ $remote_host = necriam ]]
[35]	919	then
	920
	921	if (( nodes > 1 ))
	922	then
	923	cat > $job_to_send << %%END%%
	924	#!/bin/ksh
	925	#PBS -b $nodes
	926	#PBS -o $remote_dayfile
[253]	927	#PBS -N $job_name
[35]	928	#PBS -j o
[251]	929	#PBS -v MPIPROGINV=YES
[35]	930
	931	%%END%%
	932
	933	elif [[ $numprocs != 0 ]]
	934	then
	935	cat > $job_to_send << %%END%%
	936	#!/bin/ksh
	937	#PBS -q ${queue}
	938	#PBS -o $remote_dayfile
[253]	939	#PBS -N $job_name
[35]	940	#PBS -j o
[251]	941	#PBS -v MPIPROGINV=YES
[35]	942
	943	%%END%%
	944
	945	else
	946	cat > $job_to_send << %%END%%
	947	#!/bin/ksh
	948	#PBS -o $remote_dayfile
	949	#PBS -j o
	950
	951	%%END%%
	952
	953	fi
	954
[1]	955	elif [[ $remote_host = lctit ]]
	956	then
	957	cat > $job_to_send << %%END%%
	958	#!/bin/bash
	959	#$ -S /bin/bash
	960	cd $job_catalog
[69]	961	export PATH=$PALM_BIN:\$PATH
	962	echo \$PATH
[1]	963
	964	%%END%%
	965
	966	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
[118]	967	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
[122]	968	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
[1]	969
	970	else
	971
	972	cat > $job_to_send << %%END%%
	973	# @\$-q ${queue}
	974	# @\$-l${qsubtime} $timestring
	975	# @\$-l${qsubmem} ${memory}mb
	976	# @\$-o $remote_dayfile
	977	# @\$-eo
	978
	979	%%END%%
	980
	981	fi
	982
	983
	984	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	985	# BEI EXIT VERANLASSEN
	986	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	987	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	988	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	989	then
	990	echo "set +vx" >> $job_to_send
	991	echo "trap '" >> $job_to_send
	992	echo "set +vx" >> $job_to_send
[181]	993	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]	994	then
	995	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
	996	then
	997	return_queue=c1
	998	elif [[ $remote_host = ibms ]]
	999	then
	1000	return_queue=p_normal
	1001	elif [[ $remote_host = ibmy ]]
	1002	then
	1003	return_queue=serial
[325]	1004	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
	1005	then
[326]	1006	return_queue=serialq
[1]	1007	elif [[ $remote_host = lctit ]]
	1008	then
[118]	1009	return_queue=sla3
[251]	1010	elif [[ $remote_host = necriam ]]
[1]	1011	then
[253]	1012	return_queue=SP
[1]	1013	else
	1014	return_queue=unknown
	1015	fi
	1016
	1017	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1018	then
	1019
	1020	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1021	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	1022	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	1023	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	1024	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	1025	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	1026	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
[312]	1027	if [[ $host != "ibmh" ]]
	1028	then
	1029	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	1030	fi
[1]	1031	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	1032	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	1033
	1034	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	1035	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	1036
	1037	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	1038	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
	1039	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	1040
	1041	elif [[ $remote_host = nech ]]
	1042	then
	1043	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1044	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1045	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1046	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1047	echo "#PBS -j o" >> $job_to_send
	1048	echo " " >> $job_to_send
	1049	echo "set -x" >> $job_to_send
	1050	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1051	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1052	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1053	echo "%%END%%" >> $job_to_send
	1054
[251]	1055	elif [[ $remote_host = necriam ]]
[35]	1056	then
	1057	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1058	echo "#PBS -q $return_queue" >> $job_to_send
	1059	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1060	echo "#PBS -j o" >> $job_to_send
	1061	echo " " >> $job_to_send
	1062	echo "set -x" >> $job_to_send
	1063	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1064	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1065	echo "%%END%%" >> $job_to_send
	1066
[181]	1067	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1068	then
	1069	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1070	echo "#!/bin/bash" >> $job_to_send
	1071	echo "#PBS -N job_protocol_transfer" >> $job_to_send
[325]	1072	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[181]	1073	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[366]	1074	echo "#PBS -l feature=data" >> $job_to_send
[181]	1075	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1076	echo "#PBS -j oe" >> $job_to_send
	1077	echo " " >> $job_to_send
	1078	echo ". /usr/share/modules/init/bash" >> $job_to_send
	1079	echo "set -x" >> $job_to_send
	1080	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1081	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1082	echo "%%END%%" >> $job_to_send
	1083
[1]	1084	elif [[ $remote_host = lctit ]]
	1085	then
	1086	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1087	echo "set -x" >> $job_to_send
	1088	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
	1089	echo "cd $job_catalog" >> $job_to_send
	1090	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1091	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1092	echo "%%END%%" >> $job_to_send
	1093
	1094	else
	1095
	1096	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1097	echo "# @\\\$-q $return_queue" >> $job_to_send
	1098	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1099	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1100	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1101	then
	1102	echo "# @\$-l mpp_p=0" >> $job_to_send
	1103	fi
	1104	echo '# @\$-lF 10mb' >> $job_to_send
	1105	# echo '# @\$-o /dev/null' >> $job_to_send
	1106	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1107	echo '# @\\\$-eo' >> $job_to_send
	1108	echo " " >> $job_to_send
	1109	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1110	then
	1111	echo "set +vx" >> $job_to_send
	1112	echo ". .profile" >> $job_to_send
	1113	fi
	1114	echo "set -x" >> $job_to_send
	1115	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	1116	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1117	echo "%%END%%" >> $job_to_send
	1118	fi
	1119
	1120	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1121	then
	1122	echo "llsubmit scpjob.$kennung" >> $job_to_send
[181]	1123	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1124	then
	1125	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1126	echo "msub scpjob.$kennung" >> $job_to_send
[1]	1127	elif [[ $remote_host = lctit ]]
	1128	then
	1129	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1130	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
	1131	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
	1132	echo "sleep 10" >> $job_to_send
	1133	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1134	then
	1135	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	1136	elif [[ $remote_host = t3es ]]
	1137	then
	1138	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
	1139	else
	1140	echo "qsub scpjob.$kennung" >> $job_to_send
	1141	fi
	1142	echo "rm scpjob.$kennung" >> $job_to_send
	1143	if [[ $remote_host = nech ]]
	1144	then
	1145	echo "cd -" >> $job_to_send
	1146	fi
	1147	else
	1148	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1149	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1150	fi
	1151	echo "set -x" >> $job_to_send
	1152	echo " ' exit" >> $job_to_send
	1153	echo "set -x" >> $job_to_send
	1154	fi
	1155
	1156
	1157
	1158	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	1159	cat $file_to_send >> $job_to_send
[69]	1160
[1]	1161	if [[ $remote_host = ibm ]]
	1162	then
	1163	echo " " >> $job_to_send
	1164	echo "exit" >> $job_to_send
	1165	fi
[69]	1166	if [[ $remote_host = lctit ]]
	1167	then
	1168	echo " " >> $job_to_send
	1169	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1170	fi
[1]	1171
	1172
	1173
	1174	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	1175	if [[ -z $remote_user ]]
	1176	then
	1177	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	1178	then
	1179	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	1180	fi
	1181	fi
	1182
	1183
	1184
	1185	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	1186	if [[ $no_submit = false ]]
	1187	then
	1188	if [[ $remote_host != $local_host ]]
	1189	then
	1190	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	1191	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	1192	then
[82]	1193	job_catalog_save=$job_catalog
	1194	job_catalog=job_queue
	1195	elif [[ $remote_host = nech ]]
	1196	then
	1197	job_catalog_save=$job_catalog
	1198	job_catalog=/hpf/b/${remote_user}/job_queue
	1199	fi
	1200	if [[ $local_host = decalpha ]]
	1201	then
	1202	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	1203	# PASSWORT AB
	1204	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	1205	elif [[ $remote_host = nech ]]
	1206	then
	1207	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	1208	# TRANSFERIERT WERDEN
	1209	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1210	else
[82]	1211	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1212	fi
[82]	1213	if [[ $? = 1 ]]
	1214	then
	1215	locat=scp; exit
	1216	fi
	1217	if [[ $remote_host = ibms ]]
	1218	then
	1219	job_catalog=$job_catalog_save
	1220	fi
[1]	1221	[[ $verify = true ]] && printf "\n >>> finished\n"
	1222	else
	1223	eval job_catalog=$job_catalog
	1224	cp $job_to_send ${job_catalog}/$job_on_remhost
	1225	fi
	1226
	1227
	1228
	1229	# NQS- BZW. LOADLEVELER-JOB STARTEN
	1230	if [[ $remote_host != $local_host ]]
	1231	then
	1232	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[82]	1233	if [[ $remote_host != lctit ]]
[1]	1234	then
[372]	1235	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
	1236	then
	1237	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
	1238	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
	1239	else
	1240	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
	1241	fi
[1]	1242	else
[206]	1243	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
[82]	1244	# MIT SSH, DESHALB AUFRUF PER PIPE
	1245	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
	1246	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	1247	echo "first try" > send_protocol
	1248	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
	1249	do
	1250	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
	1251	then
	1252	printf "\n trying to submit job to TIT cluster..."
	1253	else
	1254	printf "\n +++ failed ... trying to submit again ..."
	1255	fi
	1256	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
	1257	cat send_protocol
	1258	done
	1259	sleep 10
	1260	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	1261	echo "$submcom $job_on_remhost"
	1262	rm send_protocol
	1263	fi
[1]	1264
	1265	[[ $verify = true ]] && printf " >>> o.k.\n"
	1266	else
	1267	cd $job_catalog
[210]	1268	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
[1]	1269	then
	1270	eval $submcom $job_on_remhost
[367]	1271	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 \|\| $localhost = lck ]]
[108]	1272	then
	1273	eval $submcom $job_on_remhost
	1274	echo "$submcom $job_on_remhost"
	1275	chmod u+x $job_on_remhost
[292]	1276	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
	1277	then
	1278	eval $submcom $job_on_remhost
[1]	1279	elif [[ $local_host = nech ]]
	1280	then
	1281	if [[ $queue = default ]]
	1282	then
	1283	eval $submcom $job_on_remhost
	1284	else
	1285	eval $submcom -q $queue $job_on_remhost
	1286	fi
	1287	else
	1288	qsub $job_on_remhost
	1289	fi
[69]	1290	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
	1291	[[ $local_host != lctit ]] && rm $job_on_remhost
[1]	1292	cd - > /dev/null
	1293	fi
	1294	fi
	1295
	1296
	1297
	1298	# ABSCHLUSSARBEITEN
	1299
	1300	if [[ $no_submit = false ]]
	1301	then
	1302	rm $job_to_send
	1303	fi
	1304	[[ $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 |