Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 121

Last change on this file since 121 was 118, checked in by raasch, 17 years ago
preliminary adjustments for mrun and mbuild to tsubame
Property svn:keywords set to `Rev`
File size: 36.7 KB

Rev	Line
[1]	1	#!/bin/ksh
[82]	2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
[66]	3	# $Id: subjob 54 2007-03-08 00:00:02Z raasch $
[1]	4
	5	# Prozedur zum automatischen Generieren von Batch-Jobs, die unter NQS
	6	# laufen sollen und deren Ergebnis (Dayfile) zum Job-generierenden
	7	# Host zurueckgeschickt werden sollen
	8
	9
	10	# letzte Aenderung:
	11	# 29/06/94 - Siggi - Beginn mit Erstellung der Originalversion
	12	# 08/07/94 - Siggi - Originalversion abgeschlossen (Version 1.0)
	13	# 06/02/98 - Siggi - berte validiert
	14	# 27/01/01 - Siggi - ground.yonsei.ac.kr validiert, Jobs zur T3E in Korea
	15	# moeglich
	16	# 08/02/01 - Siggi - alle subjob-Meldungen ins englische uebersetzt
	17	# 25/05/02 - Siggi - Unterstuetzung des LoadLeveler
	18	# 30/05/02 - Siggi - Validierung fuer ibm-Rechner in Seoul (nobel) sowie
	19	# allgemeine Anpassungen fuer ibm-Rechner
	20	# 15/10/02 - Siggi - Neue Default-Jobklasse (p_normal) fuer IBM in Seoul
	21	# Ruecktransfer des Jobprotokolls fuer diese
	22	# Maschine abgeschaltet
	23	# 31/10/02 - Siggi - berni validiert
	24	# 06/11/02 - Siggi - Neue Jobklassen auf ibmb und ibmh
	25	# 08/11/02 - Siggi - quanero validiert
	26	# 11/12/02 - Siggi - Notification fuer Transfer-Jobs abgeschaltet
	27	# 23/01/03 - Siggi - hostname nobel changed to nobela
	28	# 06/02/03 - Siggi - gregale validated
	29	# 12/02/03 - Siggi - orkan and maestro validated
	30	# 21/02/03 - Siggi - all nobel nodes in Seoul validated
	31	# 12/03/03 - Siggi - nec at DKRZ validated
	32	# 13/03/03 - Siggi - new nqs resource variable Memory
	33	# 07/04/03 - Siggi - processor request option -c on nech needs tasks per
	34	# node
	35	# 11/04/03 - Siggi - network on ibms has attribute "shared"
	36	# 31/07/03 - Siggi - nqs2 on nech implemented (provisional: -h nech2)
	37	# cxxl added to ibmh
	38	# 29/08/03 - Siggi - changes in job queues and communication system on
	39	# ibms
	40	# 24/10/03 - Siggi - using alternate hanni address 130.75.4.2
	41	# 30/10/03 - Siggi - nech is not supported any more
	42	# 10/11/03 - Siggi - nech2 renamed to nech
	43	# 20/11/03 - Siggi - submit command on nech changed from qsub.test to qsub
	44	# 29/03/04 - Siggi - ground not supported any more, gfdl3 validated
	45	# 31/03/04 - Siggi - new option -N for node usage
	46	# 12/04/04 - Siggi - scp2 instead of scp used for transfer from decalpha
	47	# due to error in ssh installation (otherwise a prompt
	48	# for the password appears)
	49	# 23/07/04 - Siggi - changes due to the new berni configuration
	50	# (federation switch)
	51	# 01/09/04 - Gerald new job-classes on hanni
	52	# 08/09/04 - Siggi - hanni IP address changed to 130.75.4.10
	53	# 23/11/04 - Siggi - new job class cdata on hanni and berni
	54	# 03/12/04 - Siggi - notification on ibm switched of in case of
	55	# delete_dayfile = true, node usage in cdev set to
	56	# shared
	57	# 16/02/05 - Gerald hababai validated
	58	# 29/03/05 - Micha - new job class channi on hanni
	59	# 11/05/05 - Siggi - ConsumableMemory is now required as resource keyword
	60	# on ibms
	61	# 24/05/05 - Siggi - Default queue on ibms changed from p_normal_1.3 to
	62	# p_normal
	63	# 30/06/05 - Siggi - network changed for queue cdev from "us" to "ip"
	64	# 12/07/05 - Siggi - in network.mpi on ibmh/ibmb "csss" changed to
	65	# "sn_all", new job class cexp
	66	# 08/09/05 - Siggi - IP-address of gfdl3 changed
	67	# 31/10/05 - Siggi - new job class pp on hurricane, serial jobs on
	68	# hurricane (with -X0)
	69	# 01/11/05 - Siggi - missing queue for jobs submitted on nech (for nech)
	70	# added
	71	# 30/12/05 - Siggi - change of IP adresses in subnet 130.75.105
	72	# 09/02/06 - Siggi - ibmy admitted
	73	# 10/02/06 - Siggi - scp2 changed to /bin/scp on decalpha
	74	# 13/04/06 - Siggi - ostria admitted
	75	# 18/04/06 - Siggi - new option -O for OpenMP usage
	76	# 24/05/06 - Siggi - lctit admitted, ftpjob renamed scpjob
	77	# 25/07/06 - Siggi - gfdl5 (ibmy) admitted for submitting jobs
	78	# 27/09/06 - Siggi - breg/hreg extended with berni/hanni
	79	# 25/10/06 - Siggi - data_limit set to 1.76 GByte on hanni and berni
	80	# 28/11/06 - Siggi - levanto admitted
[5]	81	# 13/02/07 - Siggi - hpmuk releated code removed
[35]	82	# 01/03/07 - Siggi - adjustments for RIAM machines gate and NEC-SX8 (n-sx)
[82]	83	# 12/04/07 - Siggi - option -f (filetransfer protocol) removed, scp only
[102]	84	# 27/07/07 - Siggi - autan admitted
[108]	85	# 03/08/07 - Marcus- lcfimm admitted
[113]	86	# 08/10/07 - Siggi - further job classes added for hanni (csoloh...)
[118]	87	# 15/10/07 - Siggi - preliminary adjustments for lctit based on Jin's
	88	# suggestions
[1]	89
	90
[113]	91
[1]	92	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
	93	delete_dayfile=false
	94	locat=normal
	95	no_default_queue=none
	96	no_submit=false
	97	job_catalog="~/job_queue"
	98	job_name=none
	99	local_user=$LOGNAME
	100	node_usage=shared
	101	numprocs=1
	102	punkte="..........................................................."
	103	submcom=qsub
	104	queue=default
	105	remote_host=none
	106	remote_user=""
	107	verify=true
	108
	109	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
	110	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
	111	typeset -L20 spalte1
	112	typeset -R40 spalte2
	113	typeset -L60 spalte3
	114
	115
	116
	117	# FEHLERBEHANDLUNG
	118	# BEI EXIT:
	119	trap 'if [[ $locat != normal ]]
	120	then
	121	case $locat in
	122	(option) printf "\n --> available optios can be displayed"
	123	printf " by typing:"
	124	printf "\n \"subjob ?\" \n";;
	125	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	126	(*) printf "\n +++ unknown error"
	127	printf "\n please inform S. Raasch!\n"
	128	esac
	129	[[ -f $job_to_send ]] && rm $job_to_send
	130	printf "\n\n+++ SUBJOB killed \n\n"
	131	fi' exit
	132
	133
	134	# BEI TERMINAL-BREAK:
	135	trap '[[ -f $job_to_send ]] && rm $job_to_send
	136	printf "\n\n+++ SUBJOB killed \n\n"
	137	exit
	138	' 2
	139
	140
	141
	142
	143	# LOKALEN HOSTNAMEN FESTSTELLEN
	144	local_host=$(hostname)
	145
	146
	147
	148	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
	149	# UEBERHAUPT ZULAESSIG IST
	150	case $local_host in
	151	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[102]	152	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	153	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[1]	154	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
[12]	155	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[108]	156	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
[1]	157	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	158	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[108]	159	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
[101]	160	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[35]	161	(gate\|n-sx) local_addres=133.5.178.11; local_host=neck;;
[1]	162	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	163	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	164	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	165	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[1]	166	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
[5]	167	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
	168	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
	169	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[1]	170	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	171	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	172	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
	173	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[1]	174	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
	175	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
	176	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
	177	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
	178	(*) printf "\n +++ \"$local_host\" unknown";
	179	printf "\n please inform S. Raasch!";
	180	locat=parameter; exit;;
	181	esac
	182
	183
	184
	185	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	186	remote_host=$local_host
	187
	188
	189
	190
	191	# PROZEDUROPTIONEN EINLESEN
[82]	192	while getopts :c:dDh:m:n:N:O:q:t:T:u:vX: option
[1]	193	do
	194	case $option in
	195	(c) job_catalog=$OPTARG;;
	196	(d) delete_dayfile=true;;
	197	(D) no_submit=true;;
	198	(h) remote_host=$OPTARG;;
	199	(m) memory=$OPTARG;;
	200	(n) job_name=$OPTARG;;
	201	(N) node_usage=$OPTARG;;
	202	(O) threads_per_task=$OPTARG;;
	203	(q) no_default_queue=$OPTARG;;
	204	(t) cputime=$OPTARG;;
	205	(T) tasks_per_node=$OPTARG;;
	206	(u) remote_user=$OPTARG;;
	207	(v) verify=false;;
	208	(X) numprocs=$OPTARG;;
	209	(\?) printf "\n +++ Option $OPTARG unknown \n";
	210	locat=option; exit;;
	211	esac
	212	done
	213
	214
	215	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	216	shift OPTIND-1; file_to_send=$1
	217
	218
	219	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	220	if [ "$1" = "?" ]
	221	then
	222	(printf "\n *** subjob can be called as follows:\n"
	223	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	224	printf "\n Description of available options:\n"
	225	printf "\n Option Description Default-Value"
	226	printf "\n -c job-input- and output-catalog ~/job_queue"
	227	printf "\n -d no job-protocol will be created ---"
	228	printf "\n -D only the job-file will be created ---"
	229	printf "\n -h execution host, available hosts: $remote_host"
[5]	230	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
[35]	231	printf "\n lctit, nech, neck, unics"
[1]	232	printf "\n -m memory demand per process in MByte ---"
	233	printf "\n -n jobname <jobdatei>"
	234	printf "\n -O threads per task (for OpenMP usage) 1"
	235	printf "\n -q job-queue to be used default"
	236	printf "\n -t allowed cpu-time in seconds ---"
	237	printf "\n -T tasks per node (on parallel hosts) ---"
	238	printf "\n -u username on execution host from .netrc"
	239	printf "\n -v no prompt for confirmation ---"
	240	printf "\n -X # of processors (on parallel hosts) 1"
	241	printf "\n "
	242	printf "\n The only possible positional parameter is <jobfile>:"
	243	printf "\n The complete NQS-job must be provided here."
	244	printf "\n <jobfile>=? creates this outline\n\n") \| more
	245	exit
	246	fi
	247
	248
	249
	250	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	251	if [[ "$file_to_send" = "" ]]
	252	then
	253	printf "\n +++ job-file missing"
	254	locat=parameter; exit
	255	else
	256	if [[ -f $file_to_send ]]
	257	then
	258	true
	259	else
	260	printf "\n +++ job-file: "
	261	printf "\n $file_to_send"
	262	printf "\n does not exist"
	263	locat=parameter; exit
	264	fi
	265	fi
	266
	267
	268
	269	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	270	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	271	if [[ $job_name = none ]]
	272	then
	273	job_name=$file_to_send
	274	fi
	275	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	276	then
	277	printf "\n +++ job-file name: "
	278	printf "\n $job_name"
	279	printf "\n must not contain \"/\"-characters"
	280	locat=parameter; exit
	281	fi
	282
	283
	284
	285
	286	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	287	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	288	if [[ $remote_host = none ]]
	289	then
	290	printf "\n +++ host missing"
	291	locat=option; exit
	292	else
	293	case $remote_host in
	294	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	295	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	296	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	297	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	298	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[108]	299	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
[118]	300	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
[1]	301	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
[35]	302	(neck) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=133.5.178.11; submcom="/usr/bin/nqsII/qsub";;
[1]	303	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	304	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	305	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	306	locat=parameter; exit;;
	307	esac
	308	fi
	309
	310
	311	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	312	if [[ $no_default_queue != none ]]
	313	then
	314	error=false
	315	ndq=$no_default_queue
	316	case $remote_host in
	317	(ibm) case $ndq in
	318	(p690_express\|p690_standard\|p690_long) error=false;;
	319	(*) error=true;;
	320	esac;;
	321	(ibmb) case $ndq in
	322	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
	323	(*) error=true;;
	324	esac;;
	325	(ibmh) case $ndq in
[113]	326	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|csoloh3h\|csoloh6h\|csoloh12h\|csoloh1d\|csoloh2d\|cspec) error=false;;
[1]	327	(*) error=true;;
	328	esac;;
	329	(ibms) case $ndq in
	330	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	331	(*) error=true;;
	332	esac;;
	333	(ibmy) case $ndq in
	334	(parallel) error=false;;
	335	(*) error=true;;
	336	esac;;
	337	(lctit) case $ndq in
[69]	338	(default\|high\|bes1\|bes2\|sla1\|sla2) error=false;;
[1]	339	(*) error=true;;
	340	esac;;
	341	(t3eb) case $ndq in
	342	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	343	(*) error=true;;
	344	esac;;
	345	(t3eh) case $ndq in
	346	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	347	(*) error=true;;
	348	esac;;
	349	(t3ej2\|t3ej5) case $ndq in
	350	(low\|normal\|high) error=false;;
	351	(*) error=true;;
	352	esac;;
	353	(t3es) case $ndq in
	354	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	355	(*) error=true;;
	356	esac;;
	357	(unics) case $ndq in
	358	(unics\|ht) error=false;;
	359	(*) error=true;;
	360	esac;;
	361	esac
	362	if [[ $error = true ]]
	363	then
	364	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	365	locat=parameter; exit
	366	else
	367	queue=$no_default_queue
	368	fi
	369	fi
	370
	371
	372
	373	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
	374	if [[ $node_usage != shared && $queue = cdev ]]
	375	then
	376	node_usage=shared
	377	fi
	378
	379
	380
	381	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	382	# AUFTEILEN
	383	done=false
	384	while [[ $done = false ]]
	385	do
	386	if (( $cputime <= 0 ))
	387	then
	388	printf "\n +++ wrong cpu-time or cpu-time missing"
	389	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	390	printf "\n >>> "
	391	read cputime 1>/dev/null 2>&1
	392	else
	393	done=true
	394	fi
	395	done
	396	if [[ $remote_host = nech ]]
	397	then
	398	if (( tasks_per_node != 0 ))
	399	then
	400	(( cputime = cputime * tasks_per_node ))
	401	elif [[ $numprocs != 0 ]]
	402	then
	403	(( cputime = cputime * numprocs ))
	404	fi
	405	fi
	406	(( stunden = cputime / 3600 ))
	407	(( resttime = cputime - stunden * 3600 ))
	408	(( minuten = resttime / 60 ))
	409	(( sekunden = resttime - minuten * 60 ))
	410	timestring=${stunden}:${minuten}:${sekunden}
	411
	412
	413
	414	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	415	done=false
	416	while [[ $done = false ]]
	417	do
	418	if (( memory <= 0 ))
	419	then
	420	printf "\n +++ wrong memory demand or memory demand missing"
	421	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	422	printf "\n >>> "
	423	read memory 1>/dev/null 2>&1
	424	else
	425	done=true
	426	fi
	427	done
	428
[35]	429	if [[ $remote_host = nech \|\| $remote_host = neck ]]
[1]	430	then
	431	if (( tasks_per_node != 0 ))
	432	then
	433	(( Memory = memory * tasks_per_node / 1000 ))
	434	elif [[ $numprocs != 0 ]]
	435	then
	436	(( Memory = memory * numprocs / 1000 ))
	437	else
	438	(( Memory = memory / 1000 ))
	439	fi
	440	fi
	441
	442
	443	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	444	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	445	then
	446	(( memory = memory * threads_per_task ))
	447	fi
	448
	449
	450	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	451	if (( tasks_per_node != 0 ))
	452	then
	453	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
	454	fi
	455
	456
	457
	458	# HEADER-AUSGABE
	459	if [[ $verify = true ]]
	460	then
	461	printf "\n\n"
	462	printf "#--------------------------------------------------------------# \n"
	463	spalte1=SUBJOB;spalte2=$(date)
	464	printf "\| $spalte1$spalte2 \| \n"
	465	printf "\| \| \n"
	466	printf "\| values of parameters/options: \| \n"
	467	spalte1=local_host$punkte; spalte2=$punkte$local_host
	468	printf "\| $spalte1$spalte2 \| \n"
	469	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	470	printf "\| $spalte1$spalte2 \| \n"
	471	spalte1=queue$punkte; spalte2=$punkte$queue
	472	printf "\| $spalte1$spalte2 \| \n"
	473	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	474	printf "\| $spalte1$spalte2 \| \n"
	475	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	476	printf "\| $spalte1$spalte2 \| \n"
	477	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	478	printf "\| $spalte1$spalte2 \| \n"
	479	printf "#--------------------------------------------------------------# \n\n"
	480
	481
	482	# KONTROLLABFRAGE, OB ALLES O.K.
	483	antwort="dummy"
	484	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	485	do
	486	read antwort?" >>> continue (y/n) ? "
	487	done
	488	if [[ $antwort = n \|\| $antwort = N ]]
	489	then
	490	locat=verify; exit
	491	fi
	492	printf "\n"
	493	fi
	494
	495
	496
	497	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	498	kennung=$RANDOM
	499	job_on_remhost=${job_name}_${kennung}_$local_host
	500	job_to_send=job_to_send_$kennung
	501	if [[ $delete_dayfile = false ]]
	502	then
	503	remote_dayfile=${local_host}_${job_name}_result_$kennung
	504	local_dayfile=${remote_host}_${job_name}
	505	else
	506	remote_dayfile=/dev/null
	507	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	508	then
	509	notification='# @ notification = never'
	510	fi
	511	fi
	512
	513
	514
	515	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
	516	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	517	then
	518
	519	if [[ $remote_host = ibmy ]]
	520	then
	521	consumable_memory=""
	522	else
	523	consumable_memory="ConsumableMemory($memory mb)"
	524	fi
	525	if [[ $queue = cdev ]]
	526	then
	527	data_limit="# @ data_limit = 1.76gb"
	528	network_to_use="# @ network.mpi = sn_all,shared,ip"
	529	else
	530	if [[ $remote_host = ibms ]]
	531	then
	532	network_to_use="# @ network.mpi = csss,shared,us"
	533	elif [[ $remote_host = ibmy ]]
	534	then
	535	network_to_use=""
	536	else
	537	network_to_use="# @ network.mpi = sn_all,shared,us"
	538	data_limit="# @ data_limit = 1.76gb"
	539	fi
	540	fi
	541
	542	cat > $job_to_send << %%END%%
	543	#!/bin/ksh
	544
	545	# @ job_type = parallel
	546	# @ job_name = palm
	547	# @ wall_clock_limit = ${timestring},$timestring
	548	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	549	# @ output = $remote_dayfile
	550	# @ error = $remote_dayfile
	551	# @ image_size = 50
	552	# @ class = $queue
	553	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
	554	$network_to_use
	555	$data_limit
	556	$notification
	557
	558	%%END%%
	559
	560	if (( nodes > 0 ))
	561	then
	562
	563	cat >> $job_to_send << %%END%%
	564	# @ node = $nodes
	565	# @ tasks_per_node = $tasks_per_node
	566	# @ node_usage = $node_usage
	567	# @ queue
	568
	569	%%END%%
	570
	571	else
	572
	573	if [[ $remote_host != ibmy ]]
	574	then
	575
	576	cat >> $job_to_send << %%END%%
	577	# @ blocking = unlimited
	578	# @ total_tasks = $numprocs
	579	# @ node_usage = $node_usage
	580	# @ queue
	581
	582	%%END%%
	583
	584	else
	585
	586	cat >> $job_to_send << %%END%%
	587	# @ node = 1
	588	# @ total_tasks = $numprocs
	589	# @ queue
	590
	591	%%END%%
	592
	593	fi
	594
	595	fi
	596
	597	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	598	then
	599
	600	cat > $job_to_send << %%END%%
	601	#!/bin/ksh
	602
	603	# @ job_type = serial
	604	# @ node_usage = $node_usage
	605	# @ job_name = palm
	606	# @ wall_clock_limit = ${timestring},$timestring
	607	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	608	# @ output = $remote_dayfile
	609	# @ error = $remote_dayfile
	610	# @ class = $queue
	611	$notification
	612
	613	# @ queue
	614
	615	%%END%%
	616
[108]	617	elif [[ $remote_host = lcfimm ]]
	618	then
	619
	620	if [[ $numprocs != 0 ]]
	621	then
	622	cat > $job_to_send << %%END%%
	623	#!/bin/ksh
	624	#PBS -N $job_name
	625	#PBS -A nersc
	626	#PBS -l walltime=$timestring
	627	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
	628	#PBS -l pmem=${memory}mb
	629	#PBS -m abe
	630	#PBS -M igore@nersc.no
	631	#PBS -o $remote_dayfile
	632	#PBS -j oe
	633	mpd &
	634
	635	%%END%%
	636
	637	else
	638	cat > $job_to_send << %%END%%
	639	#!/bin/ksh
	640	#PBS -N $job_name
	641	#PBS -A nersc
	642	#PBS -l walltime=$timestring
	643	#PBS -l ncpus=1
	644	#PBS -l pmem=${memory}mb
	645	#PBS -m abe
	646	#PBS -M igore@nersc.no
	647	#PBS -o $remote_dayfile
	648	#PBS -j oe
	649
	650	%%END%%
	651
	652	fi
	653
[1]	654	elif [[ $remote_host = nech ]]
	655	then
	656
	657	if (( nodes > 1 ))
	658	then
	659	cat > $job_to_send << %%END%%
	660	#!/bin/ksh
	661	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	662	#PBS -l ${qsubmem}=${Memory}gb
	663	#PBS -b $nodes
	664	#PBS -o $remote_dayfile
	665	#PBS -N palm
	666	#PBS -j o
	667	#PBS -T mpisx
	668
	669	%%END%%
	670
	671	elif [[ $numprocs != 0 ]]
	672	then
	673	cat > $job_to_send << %%END%%
	674	#!/bin/ksh
	675	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	676	#PBS -l ${qsubmem}=${Memory}gb
	677	#PBS -o $remote_dayfile
	678	#PBS -N palm
	679	#PBS -j o
	680
	681	%%END%%
	682
	683	else
	684	cat > $job_to_send << %%END%%
	685	#!/bin/ksh
	686	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	687	#PBS -o $remote_dayfile
	688	#PBS -j o
	689
	690	%%END%%
	691
	692	fi
	693
[35]	694	elif [[ $remote_host = neck ]]
	695	then
	696
	697	if (( nodes > 1 ))
	698	then
	699	cat > $job_to_send << %%END%%
	700	#!/bin/ksh
	701	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	702	#PBS -l ${qsubmem}=${Memory}gb
	703	#PBS -b $nodes
	704	#PBS -o $remote_dayfile
	705	#PBS -N palm
	706	#PBS -j o
	707	#PBS -T mpisx
	708
	709	%%END%%
	710
	711	elif [[ $numprocs != 0 ]]
	712	then
	713	cat > $job_to_send << %%END%%
	714	#!/bin/ksh
	715	#PBS -q ${queue}
	716	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	717	#PBS -l ${qsubmem}=${Memory}gb
	718	#PBS -o $remote_dayfile
	719	#PBS -N palm
	720	#PBS -j o
	721	#PBS -T mpisx
	722
	723	%%END%%
	724
	725	else
	726	cat > $job_to_send << %%END%%
	727	#!/bin/ksh
	728	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	729	#PBS -o $remote_dayfile
	730	#PBS -j o
	731
	732	%%END%%
	733
	734	fi
	735
[1]	736	elif [[ $remote_host = lctit ]]
	737	then
	738	cat > $job_to_send << %%END%%
	739	#!/bin/bash
	740	#$ -S /bin/bash
	741	cd $job_catalog
[69]	742	export PATH=$PALM_BIN:\$PATH
	743	echo \$PATH
[1]	744
	745	%%END%%
	746
	747	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
[118]	748	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
	749	submcom="$submcom -g 4SO7O143 -N $job_name -rt 4320 -sgeout $remote_dayfile -q sla3"
[1]	750
	751	else
	752
	753	cat > $job_to_send << %%END%%
	754	# @\$-q ${queue}
	755	# @\$-l${qsubtime} $timestring
	756	# @\$-l${qsubmem} ${memory}mb
	757	# @\$-o $remote_dayfile
	758	# @\$-eo
	759
	760	%%END%%
	761
	762	fi
	763
	764
	765	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	766	# BEI EXIT VERANLASSEN
	767	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	768	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	769	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	770	then
	771	echo "set +vx" >> $job_to_send
	772	echo "trap '" >> $job_to_send
	773	echo "set +vx" >> $job_to_send
[35]	774	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
[1]	775	then
	776	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
	777	then
	778	return_queue=c1
	779	elif [[ $remote_host = ibms ]]
	780	then
	781	return_queue=p_normal
	782	elif [[ $remote_host = ibmy ]]
	783	then
	784	return_queue=serial
	785	elif [[ $remote_host = lctit ]]
	786	then
[118]	787	return_queue=sla3
[35]	788	elif [[ $remote_host = neck ]]
[1]	789	then
[35]	790	return_queue=S
[1]	791	else
	792	return_queue=unknown
	793	fi
	794
	795	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	796	then
	797
	798	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	799	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	800	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	801	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	802	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	803	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	804	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	805	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	806	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	807	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	808
	809	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	810	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	811
	812	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	813	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
	814	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	815
	816	elif [[ $remote_host = nech ]]
	817	then
	818	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	819	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	820	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	821	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	822	echo "#PBS -j o" >> $job_to_send
	823	echo " " >> $job_to_send
	824	echo "set -x" >> $job_to_send
	825	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	826	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	827	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	828	echo "%%END%%" >> $job_to_send
	829
[35]	830	elif [[ $remote_host = neck ]]
	831	then
	832	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	833	echo "#PBS -q $return_queue" >> $job_to_send
	834	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	835	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	836	echo "#PBS -j o" >> $job_to_send
	837	echo " " >> $job_to_send
	838	echo "set -x" >> $job_to_send
	839	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	840	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	841	echo "%%END%%" >> $job_to_send
	842
[1]	843	elif [[ $remote_host = lctit ]]
	844	then
	845	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	846	echo "set -x" >> $job_to_send
	847	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
	848	echo "cd $job_catalog" >> $job_to_send
	849	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	850	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	851	echo "%%END%%" >> $job_to_send
	852
	853	else
	854
	855	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	856	echo "# @\\\$-q $return_queue" >> $job_to_send
	857	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	858	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	859	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	860	then
	861	echo "# @\$-l mpp_p=0" >> $job_to_send
	862	fi
	863	echo '# @\$-lF 10mb' >> $job_to_send
	864	# echo '# @\$-o /dev/null' >> $job_to_send
	865	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	866	echo '# @\\\$-eo' >> $job_to_send
	867	echo " " >> $job_to_send
	868	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	869	then
	870	echo "set +vx" >> $job_to_send
	871	echo ". .profile" >> $job_to_send
	872	fi
	873	echo "set -x" >> $job_to_send
	874	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	875	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	876	echo "%%END%%" >> $job_to_send
	877	fi
	878
	879	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	880	then
	881	echo "llsubmit scpjob.$kennung" >> $job_to_send
	882	elif [[ $remote_host = lctit ]]
	883	then
	884	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	885	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
	886	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
	887	echo "sleep 10" >> $job_to_send
	888	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	889	then
	890	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	891	elif [[ $remote_host = t3es ]]
	892	then
	893	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
	894	else
	895	echo "qsub scpjob.$kennung" >> $job_to_send
	896	fi
	897	echo "rm scpjob.$kennung" >> $job_to_send
	898	if [[ $remote_host = nech ]]
	899	then
	900	echo "cd -" >> $job_to_send
	901	fi
	902	else
	903	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	904	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	905	fi
	906	echo "set -x" >> $job_to_send
	907	echo " ' exit" >> $job_to_send
	908	echo "set -x" >> $job_to_send
	909	fi
	910
	911
	912
	913	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	914	cat $file_to_send >> $job_to_send
[69]	915
[1]	916	if [[ $remote_host = ibm ]]
	917	then
	918	echo " " >> $job_to_send
	919	echo "exit" >> $job_to_send
	920	fi
[69]	921	if [[ $remote_host = lctit ]]
	922	then
	923	echo " " >> $job_to_send
	924	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	925	fi
[1]	926
	927
	928
	929	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	930	if [[ -z $remote_user ]]
	931	then
	932	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	933	then
	934	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	935	fi
	936	fi
	937
	938
	939
	940	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	941	if [[ $no_submit = false ]]
	942	then
	943	if [[ $remote_host != $local_host ]]
	944	then
	945	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	946	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	947	then
[82]	948	job_catalog_save=$job_catalog
	949	job_catalog=job_queue
	950	elif [[ $remote_host = nech ]]
	951	then
	952	job_catalog_save=$job_catalog
	953	job_catalog=/hpf/b/${remote_user}/job_queue
	954	fi
	955	if [[ $local_host = decalpha ]]
	956	then
	957	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	958	# PASSWORT AB
	959	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	960	elif [[ $remote_host = nech ]]
	961	then
	962	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	963	# TRANSFERIERT WERDEN
	964	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	965	else
[82]	966	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	967	fi
[82]	968	if [[ $? = 1 ]]
	969	then
	970	locat=scp; exit
	971	fi
	972	if [[ $remote_host = ibms ]]
	973	then
	974	job_catalog=$job_catalog_save
	975	fi
[1]	976	[[ $verify = true ]] && printf "\n >>> finished\n"
	977	else
	978	eval job_catalog=$job_catalog
	979	cp $job_to_send ${job_catalog}/$job_on_remhost
	980	fi
	981
	982
	983
	984	# NQS- BZW. LOADLEVELER-JOB STARTEN
	985	if [[ $remote_host != $local_host ]]
	986	then
	987	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[82]	988	if [[ $remote_host != lctit ]]
[1]	989	then
[82]	990	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[1]	991	else
[82]	992	# TIT ERLAUBT NUR DIE AUSFÜHRUNG GANZ BESTIMMTER KOMMANDOS
	993	# MIT SSH, DESHALB AUFRUF PER PIPE
	994	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
	995	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	996	echo "first try" > send_protocol
	997	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
	998	do
	999	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
	1000	then
	1001	printf "\n trying to submit job to TIT cluster..."
	1002	else
	1003	printf "\n +++ failed ... trying to submit again ..."
	1004	fi
	1005	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
	1006	cat send_protocol
	1007	done
	1008	sleep 10
	1009	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	1010	echo "$submcom $job_on_remhost"
	1011	rm send_protocol
	1012	fi
[1]	1013
	1014	[[ $verify = true ]] && printf " >>> o.k.\n"
	1015	else
	1016	cd $job_catalog
	1017	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
	1018	then
	1019	eval $submcom $job_on_remhost
[108]	1020	elif [[ $local_host = lcfimm ]]
	1021	then
	1022	eval $submcom $job_on_remhost
	1023	echo "$submcom $job_on_remhost"
	1024	chmod u+x $job_on_remhost
[1]	1025	elif [[ $local_host = lctit ]]
	1026	then
	1027	eval $submcom $job_on_remhost
	1028	echo "$submcom $job_on_remhost"
	1029	chmod u+x $job_on_remhost
	1030	elif [[ $local_host = nech ]]
	1031	then
	1032	if [[ $queue = default ]]
	1033	then
	1034	eval $submcom $job_on_remhost
	1035	else
	1036	eval $submcom -q $queue $job_on_remhost
	1037	fi
	1038	else
	1039	qsub $job_on_remhost
	1040	fi
[69]	1041	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
	1042	[[ $local_host != lctit ]] && rm $job_on_remhost
[1]	1043	cd - > /dev/null
	1044	fi
	1045	fi
	1046
	1047
	1048
	1049	# ABSCHLUSSARBEITEN
	1050
	1051	if [[ $no_submit = false ]]
	1052	then
	1053	rm $job_to_send
	1054	fi
	1055	[[ $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 |