Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 98

Last change on this file since 98 was 82, checked in by raasch, 18 years ago
vorlaeufige Standalone-Version fuer Linux-Cluster
Property svn:keywords set to `Rev`
File size: 35.2 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
[1]	84
	85
	86	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
	87	delete_dayfile=false
	88	locat=normal
	89	no_default_queue=none
	90	no_submit=false
	91	job_catalog="~/job_queue"
	92	job_name=none
	93	local_user=$LOGNAME
	94	node_usage=shared
	95	numprocs=1
	96	punkte="..........................................................."
	97	submcom=qsub
	98	queue=default
	99	remote_host=none
	100	remote_user=""
	101	verify=true
	102
	103	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
	104	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
	105	typeset -L20 spalte1
	106	typeset -R40 spalte2
	107	typeset -L60 spalte3
	108
	109
	110
	111	# FEHLERBEHANDLUNG
	112	# BEI EXIT:
	113	trap 'if [[ $locat != normal ]]
	114	then
	115	case $locat in
	116	(option) printf "\n --> available optios can be displayed"
	117	printf " by typing:"
	118	printf "\n \"subjob ?\" \n";;
	119	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	120	(*) printf "\n +++ unknown error"
	121	printf "\n please inform S. Raasch!\n"
	122	esac
	123	[[ -f $job_to_send ]] && rm $job_to_send
	124	printf "\n\n+++ SUBJOB killed \n\n"
	125	fi' exit
	126
	127
	128	# BEI TERMINAL-BREAK:
	129	trap '[[ -f $job_to_send ]] && rm $job_to_send
	130	printf "\n\n+++ SUBJOB killed \n\n"
	131	exit
	132	' 2
	133
	134
	135
	136
	137	# LOKALEN HOSTNAMEN FESTSTELLEN
	138	local_host=$(hostname)
	139
	140
	141
	142	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
	143	# UEBERHAUPT ZULAESSIG IST
	144	case $local_host in
	145	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[5]	146	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[1]	147	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
[12]	148	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[1]	149	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	150	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[35]	151	(gate\|n-sx) local_addres=133.5.178.11; local_host=neck;;
[1]	152	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	153	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	154	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	155	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[1]	156	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
[5]	157	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
	158	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
	159	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[1]	160	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	161	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	162	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
	163	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[1]	164	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
	165	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
	166	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
	167	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
	168	(*) printf "\n +++ \"$local_host\" unknown";
	169	printf "\n please inform S. Raasch!";
	170	locat=parameter; exit;;
	171	esac
	172
	173
	174
	175	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	176	remote_host=$local_host
	177
	178
	179
	180
	181	# PROZEDUROPTIONEN EINLESEN
[82]	182	while getopts :c:dDh:m:n:N:O:q:t:T:u:vX: option
[1]	183	do
	184	case $option in
	185	(c) job_catalog=$OPTARG;;
	186	(d) delete_dayfile=true;;
	187	(D) no_submit=true;;
	188	(h) remote_host=$OPTARG;;
	189	(m) memory=$OPTARG;;
	190	(n) job_name=$OPTARG;;
	191	(N) node_usage=$OPTARG;;
	192	(O) threads_per_task=$OPTARG;;
	193	(q) no_default_queue=$OPTARG;;
	194	(t) cputime=$OPTARG;;
	195	(T) tasks_per_node=$OPTARG;;
	196	(u) remote_user=$OPTARG;;
	197	(v) verify=false;;
	198	(X) numprocs=$OPTARG;;
	199	(\?) printf "\n +++ Option $OPTARG unknown \n";
	200	locat=option; exit;;
	201	esac
	202	done
	203
	204
	205	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	206	shift OPTIND-1; file_to_send=$1
	207
	208
	209	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	210	if [ "$1" = "?" ]
	211	then
	212	(printf "\n *** subjob can be called as follows:\n"
	213	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	214	printf "\n Description of available options:\n"
	215	printf "\n Option Description Default-Value"
	216	printf "\n -c job-input- and output-catalog ~/job_queue"
	217	printf "\n -d no job-protocol will be created ---"
	218	printf "\n -D only the job-file will be created ---"
	219	printf "\n -h execution host, available hosts: $remote_host"
[5]	220	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
[35]	221	printf "\n lctit, nech, neck, unics"
[1]	222	printf "\n -m memory demand per process in MByte ---"
	223	printf "\n -n jobname <jobdatei>"
	224	printf "\n -O threads per task (for OpenMP usage) 1"
	225	printf "\n -q job-queue to be used default"
	226	printf "\n -t allowed cpu-time in seconds ---"
	227	printf "\n -T tasks per node (on parallel hosts) ---"
	228	printf "\n -u username on execution host from .netrc"
	229	printf "\n -v no prompt for confirmation ---"
	230	printf "\n -X # of processors (on parallel hosts) 1"
	231	printf "\n "
	232	printf "\n The only possible positional parameter is <jobfile>:"
	233	printf "\n The complete NQS-job must be provided here."
	234	printf "\n <jobfile>=? creates this outline\n\n") \| more
	235	exit
	236	fi
	237
	238
	239
	240	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	241	if [[ "$file_to_send" = "" ]]
	242	then
	243	printf "\n +++ job-file missing"
	244	locat=parameter; exit
	245	else
	246	if [[ -f $file_to_send ]]
	247	then
	248	true
	249	else
	250	printf "\n +++ job-file: "
	251	printf "\n $file_to_send"
	252	printf "\n does not exist"
	253	locat=parameter; exit
	254	fi
	255	fi
	256
	257
	258
	259	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	260	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	261	if [[ $job_name = none ]]
	262	then
	263	job_name=$file_to_send
	264	fi
	265	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	266	then
	267	printf "\n +++ job-file name: "
	268	printf "\n $job_name"
	269	printf "\n must not contain \"/\"-characters"
	270	locat=parameter; exit
	271	fi
	272
	273
	274
	275
	276	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	277	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	278	if [[ $remote_host = none ]]
	279	then
	280	printf "\n +++ host missing"
	281	locat=option; exit
	282	else
	283	case $remote_host in
	284	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	285	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	286	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	287	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	288	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[69]	289	(lctit) queue=default; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
[1]	290	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
[35]	291	(neck) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=133.5.178.11; submcom="/usr/bin/nqsII/qsub";;
[1]	292	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	293	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	294	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	295	locat=parameter; exit;;
	296	esac
	297	fi
	298
	299
	300	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	301	if [[ $no_default_queue != none ]]
	302	then
	303	error=false
	304	ndq=$no_default_queue
	305	case $remote_host in
	306	(ibm) case $ndq in
	307	(p690_express\|p690_standard\|p690_long) error=false;;
	308	(*) error=true;;
	309	esac;;
	310	(ibmb) case $ndq in
	311	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
	312	(*) error=true;;
	313	esac;;
	314	(ibmh) case $ndq in
	315	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|cspec) error=false;;
	316	(*) error=true;;
	317	esac;;
	318	(ibms) case $ndq in
	319	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	320	(*) error=true;;
	321	esac;;
	322	(ibmy) case $ndq in
	323	(parallel) error=false;;
	324	(*) error=true;;
	325	esac;;
	326	(lctit) case $ndq in
[69]	327	(default\|high\|bes1\|bes2\|sla1\|sla2) error=false;;
[1]	328	(*) error=true;;
	329	esac;;
	330	(t3eb) case $ndq in
	331	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	332	(*) error=true;;
	333	esac;;
	334	(t3eh) case $ndq in
	335	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	336	(*) error=true;;
	337	esac;;
	338	(t3ej2\|t3ej5) case $ndq in
	339	(low\|normal\|high) error=false;;
	340	(*) error=true;;
	341	esac;;
	342	(t3es) case $ndq in
	343	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	344	(*) error=true;;
	345	esac;;
	346	(unics) case $ndq in
	347	(unics\|ht) error=false;;
	348	(*) error=true;;
	349	esac;;
	350	esac
	351	if [[ $error = true ]]
	352	then
	353	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	354	locat=parameter; exit
	355	else
	356	queue=$no_default_queue
	357	fi
	358	fi
	359
	360
	361
	362	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
	363	if [[ $node_usage != shared && $queue = cdev ]]
	364	then
	365	node_usage=shared
	366	fi
	367
	368
	369
	370	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	371	# AUFTEILEN
	372	done=false
	373	while [[ $done = false ]]
	374	do
	375	if (( $cputime <= 0 ))
	376	then
	377	printf "\n +++ wrong cpu-time or cpu-time missing"
	378	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	379	printf "\n >>> "
	380	read cputime 1>/dev/null 2>&1
	381	else
	382	done=true
	383	fi
	384	done
	385	if [[ $remote_host = nech ]]
	386	then
	387	if (( tasks_per_node != 0 ))
	388	then
	389	(( cputime = cputime * tasks_per_node ))
	390	elif [[ $numprocs != 0 ]]
	391	then
	392	(( cputime = cputime * numprocs ))
	393	fi
	394	fi
	395	(( stunden = cputime / 3600 ))
	396	(( resttime = cputime - stunden * 3600 ))
	397	(( minuten = resttime / 60 ))
	398	(( sekunden = resttime - minuten * 60 ))
	399	timestring=${stunden}:${minuten}:${sekunden}
	400
	401
	402
	403	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	404	done=false
	405	while [[ $done = false ]]
	406	do
	407	if (( memory <= 0 ))
	408	then
	409	printf "\n +++ wrong memory demand or memory demand missing"
	410	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	411	printf "\n >>> "
	412	read memory 1>/dev/null 2>&1
	413	else
	414	done=true
	415	fi
	416	done
	417
[35]	418	if [[ $remote_host = nech \|\| $remote_host = neck ]]
[1]	419	then
	420	if (( tasks_per_node != 0 ))
	421	then
	422	(( Memory = memory * tasks_per_node / 1000 ))
	423	elif [[ $numprocs != 0 ]]
	424	then
	425	(( Memory = memory * numprocs / 1000 ))
	426	else
	427	(( Memory = memory / 1000 ))
	428	fi
	429	fi
	430
	431
	432	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	433	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	434	then
	435	(( memory = memory * threads_per_task ))
	436	fi
	437
	438
	439	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	440	if (( tasks_per_node != 0 ))
	441	then
	442	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
	443	fi
	444
	445
	446
	447	# HEADER-AUSGABE
	448	if [[ $verify = true ]]
	449	then
	450	printf "\n\n"
	451	printf "#--------------------------------------------------------------# \n"
	452	spalte1=SUBJOB;spalte2=$(date)
	453	printf "\| $spalte1$spalte2 \| \n"
	454	printf "\| \| \n"
	455	printf "\| values of parameters/options: \| \n"
	456	spalte1=local_host$punkte; spalte2=$punkte$local_host
	457	printf "\| $spalte1$spalte2 \| \n"
	458	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	459	printf "\| $spalte1$spalte2 \| \n"
	460	spalte1=queue$punkte; spalte2=$punkte$queue
	461	printf "\| $spalte1$spalte2 \| \n"
	462	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	463	printf "\| $spalte1$spalte2 \| \n"
	464	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	465	printf "\| $spalte1$spalte2 \| \n"
	466	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	467	printf "\| $spalte1$spalte2 \| \n"
	468	printf "#--------------------------------------------------------------# \n\n"
	469
	470
	471	# KONTROLLABFRAGE, OB ALLES O.K.
	472	antwort="dummy"
	473	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	474	do
	475	read antwort?" >>> continue (y/n) ? "
	476	done
	477	if [[ $antwort = n \|\| $antwort = N ]]
	478	then
	479	locat=verify; exit
	480	fi
	481	printf "\n"
	482	fi
	483
	484
	485
	486	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	487	kennung=$RANDOM
	488	job_on_remhost=${job_name}_${kennung}_$local_host
	489	job_to_send=job_to_send_$kennung
	490	if [[ $delete_dayfile = false ]]
	491	then
	492	remote_dayfile=${local_host}_${job_name}_result_$kennung
	493	local_dayfile=${remote_host}_${job_name}
	494	else
	495	remote_dayfile=/dev/null
	496	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	497	then
	498	notification='# @ notification = never'
	499	fi
	500	fi
	501
	502
	503
	504	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
	505	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	506	then
	507
	508	if [[ $remote_host = ibmy ]]
	509	then
	510	consumable_memory=""
	511	else
	512	consumable_memory="ConsumableMemory($memory mb)"
	513	fi
	514	if [[ $queue = cdev ]]
	515	then
	516	data_limit="# @ data_limit = 1.76gb"
	517	network_to_use="# @ network.mpi = sn_all,shared,ip"
	518	else
	519	if [[ $remote_host = ibms ]]
	520	then
	521	network_to_use="# @ network.mpi = csss,shared,us"
	522	elif [[ $remote_host = ibmy ]]
	523	then
	524	network_to_use=""
	525	else
	526	network_to_use="# @ network.mpi = sn_all,shared,us"
	527	data_limit="# @ data_limit = 1.76gb"
	528	fi
	529	fi
	530
	531	cat > $job_to_send << %%END%%
	532	#!/bin/ksh
	533
	534	# @ job_type = parallel
	535	# @ job_name = palm
	536	# @ wall_clock_limit = ${timestring},$timestring
	537	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	538	# @ output = $remote_dayfile
	539	# @ error = $remote_dayfile
	540	# @ image_size = 50
	541	# @ class = $queue
	542	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
	543	$network_to_use
	544	$data_limit
	545	$notification
	546
	547	%%END%%
	548
	549	if (( nodes > 0 ))
	550	then
	551
	552	cat >> $job_to_send << %%END%%
	553	# @ node = $nodes
	554	# @ tasks_per_node = $tasks_per_node
	555	# @ node_usage = $node_usage
	556	# @ queue
	557
	558	%%END%%
	559
	560	else
	561
	562	if [[ $remote_host != ibmy ]]
	563	then
	564
	565	cat >> $job_to_send << %%END%%
	566	# @ blocking = unlimited
	567	# @ total_tasks = $numprocs
	568	# @ node_usage = $node_usage
	569	# @ queue
	570
	571	%%END%%
	572
	573	else
	574
	575	cat >> $job_to_send << %%END%%
	576	# @ node = 1
	577	# @ total_tasks = $numprocs
	578	# @ queue
	579
	580	%%END%%
	581
	582	fi
	583
	584	fi
	585
	586	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	587	then
	588
	589	cat > $job_to_send << %%END%%
	590	#!/bin/ksh
	591
	592	# @ job_type = serial
	593	# @ node_usage = $node_usage
	594	# @ job_name = palm
	595	# @ wall_clock_limit = ${timestring},$timestring
	596	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	597	# @ output = $remote_dayfile
	598	# @ error = $remote_dayfile
	599	# @ class = $queue
	600	$notification
	601
	602	# @ queue
	603
	604	%%END%%
	605
	606	elif [[ $remote_host = nech ]]
	607	then
	608
	609	if (( nodes > 1 ))
	610	then
	611	cat > $job_to_send << %%END%%
	612	#!/bin/ksh
	613	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	614	#PBS -l ${qsubmem}=${Memory}gb
	615	#PBS -b $nodes
	616	#PBS -o $remote_dayfile
	617	#PBS -N palm
	618	#PBS -j o
	619	#PBS -T mpisx
	620
	621	%%END%%
	622
	623	elif [[ $numprocs != 0 ]]
	624	then
	625	cat > $job_to_send << %%END%%
	626	#!/bin/ksh
	627	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	628	#PBS -l ${qsubmem}=${Memory}gb
	629	#PBS -o $remote_dayfile
	630	#PBS -N palm
	631	#PBS -j o
	632
	633	%%END%%
	634
	635	else
	636	cat > $job_to_send << %%END%%
	637	#!/bin/ksh
	638	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	639	#PBS -o $remote_dayfile
	640	#PBS -j o
	641
	642	%%END%%
	643
	644	fi
	645
[35]	646	elif [[ $remote_host = neck ]]
	647	then
	648
	649	if (( nodes > 1 ))
	650	then
	651	cat > $job_to_send << %%END%%
	652	#!/bin/ksh
	653	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	654	#PBS -l ${qsubmem}=${Memory}gb
	655	#PBS -b $nodes
	656	#PBS -o $remote_dayfile
	657	#PBS -N palm
	658	#PBS -j o
	659	#PBS -T mpisx
	660
	661	%%END%%
	662
	663	elif [[ $numprocs != 0 ]]
	664	then
	665	cat > $job_to_send << %%END%%
	666	#!/bin/ksh
	667	#PBS -q ${queue}
	668	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	669	#PBS -l ${qsubmem}=${Memory}gb
	670	#PBS -o $remote_dayfile
	671	#PBS -N palm
	672	#PBS -j o
	673	#PBS -T mpisx
	674
	675	%%END%%
	676
	677	else
	678	cat > $job_to_send << %%END%%
	679	#!/bin/ksh
	680	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	681	#PBS -o $remote_dayfile
	682	#PBS -j o
	683
	684	%%END%%
	685
	686	fi
	687
[1]	688	elif [[ $remote_host = lctit ]]
	689	then
	690	cat > $job_to_send << %%END%%
	691	#!/bin/bash
	692	#$ -S /bin/bash
	693	cd $job_catalog
[69]	694	export PATH=$PALM_BIN:\$PATH
	695	echo \$PATH
[1]	696
	697	%%END%%
	698
	699	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
	700	submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
	701
	702	else
	703
	704	cat > $job_to_send << %%END%%
	705	# @\$-q ${queue}
	706	# @\$-l${qsubtime} $timestring
	707	# @\$-l${qsubmem} ${memory}mb
	708	# @\$-o $remote_dayfile
	709	# @\$-eo
	710
	711	%%END%%
	712
	713	fi
	714
	715
	716	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	717	# BEI EXIT VERANLASSEN
	718	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	719	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	720	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	721	then
	722	echo "set +vx" >> $job_to_send
	723	echo "trap '" >> $job_to_send
	724	echo "set +vx" >> $job_to_send
[35]	725	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
[1]	726	then
	727	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
	728	then
	729	return_queue=c1
	730	elif [[ $remote_host = ibms ]]
	731	then
	732	return_queue=p_normal
	733	elif [[ $remote_host = ibmy ]]
	734	then
	735	return_queue=serial
	736	elif [[ $remote_host = lctit ]]
	737	then
	738	return_queue=default
[35]	739	elif [[ $remote_host = neck ]]
[1]	740	then
[35]	741	return_queue=S
[1]	742	else
	743	return_queue=unknown
	744	fi
	745
	746	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	747	then
	748
	749	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	750	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	751	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	752	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	753	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	754	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	755	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	756	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	757	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	758	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	759
	760	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	761	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	762
	763	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	764	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
	765	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	766
	767	elif [[ $remote_host = nech ]]
	768	then
	769	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	770	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	771	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	772	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	773	echo "#PBS -j o" >> $job_to_send
	774	echo " " >> $job_to_send
	775	echo "set -x" >> $job_to_send
	776	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	777	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	778	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	779	echo "%%END%%" >> $job_to_send
	780
[35]	781	elif [[ $remote_host = neck ]]
	782	then
	783	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	784	echo "#PBS -q $return_queue" >> $job_to_send
	785	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	786	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	787	echo "#PBS -j o" >> $job_to_send
	788	echo " " >> $job_to_send
	789	echo "set -x" >> $job_to_send
	790	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	791	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	792	echo "%%END%%" >> $job_to_send
	793
[1]	794	elif [[ $remote_host = lctit ]]
	795	then
	796	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	797	echo "set -x" >> $job_to_send
	798	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
	799	echo "cd $job_catalog" >> $job_to_send
	800	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	801	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	802	echo "%%END%%" >> $job_to_send
	803
	804	else
	805
	806	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	807	echo "# @\\\$-q $return_queue" >> $job_to_send
	808	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	809	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	810	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	811	then
	812	echo "# @\$-l mpp_p=0" >> $job_to_send
	813	fi
	814	echo '# @\$-lF 10mb' >> $job_to_send
	815	# echo '# @\$-o /dev/null' >> $job_to_send
	816	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	817	echo '# @\\\$-eo' >> $job_to_send
	818	echo " " >> $job_to_send
	819	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	820	then
	821	echo "set +vx" >> $job_to_send
	822	echo ". .profile" >> $job_to_send
	823	fi
	824	echo "set -x" >> $job_to_send
	825	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	826	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	827	echo "%%END%%" >> $job_to_send
	828	fi
	829
	830	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	831	then
	832	echo "llsubmit scpjob.$kennung" >> $job_to_send
	833	elif [[ $remote_host = lctit ]]
	834	then
	835	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	836	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
	837	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
	838	echo "sleep 10" >> $job_to_send
	839	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	840	then
	841	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	842	elif [[ $remote_host = t3es ]]
	843	then
	844	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
	845	else
	846	echo "qsub scpjob.$kennung" >> $job_to_send
	847	fi
	848	echo "rm scpjob.$kennung" >> $job_to_send
	849	if [[ $remote_host = nech ]]
	850	then
	851	echo "cd -" >> $job_to_send
	852	fi
	853	else
	854	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	855	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	856	fi
	857	echo "set -x" >> $job_to_send
	858	echo " ' exit" >> $job_to_send
	859	echo "set -x" >> $job_to_send
	860	fi
	861
	862
	863
	864	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	865	cat $file_to_send >> $job_to_send
[69]	866
[1]	867	if [[ $remote_host = ibm ]]
	868	then
	869	echo " " >> $job_to_send
	870	echo "exit" >> $job_to_send
	871	fi
[69]	872	if [[ $remote_host = lctit ]]
	873	then
	874	echo " " >> $job_to_send
	875	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	876	fi
[1]	877
	878
	879
	880	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	881	if [[ -z $remote_user ]]
	882	then
	883	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	884	then
	885	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	886	fi
	887	fi
	888
	889
	890
	891	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	892	if [[ $no_submit = false ]]
	893	then
	894	if [[ $remote_host != $local_host ]]
	895	then
	896	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	897	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	898	then
[82]	899	job_catalog_save=$job_catalog
	900	job_catalog=job_queue
	901	elif [[ $remote_host = nech ]]
	902	then
	903	job_catalog_save=$job_catalog
	904	job_catalog=/hpf/b/${remote_user}/job_queue
	905	fi
	906	if [[ $local_host = decalpha ]]
	907	then
	908	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	909	# PASSWORT AB
	910	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	911	elif [[ $remote_host = nech ]]
	912	then
	913	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	914	# TRANSFERIERT WERDEN
	915	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	916	else
[82]	917	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	918	fi
[82]	919	if [[ $? = 1 ]]
	920	then
	921	locat=scp; exit
	922	fi
	923	if [[ $remote_host = ibms ]]
	924	then
	925	job_catalog=$job_catalog_save
	926	fi
[1]	927	[[ $verify = true ]] && printf "\n >>> finished\n"
	928	else
	929	eval job_catalog=$job_catalog
	930	cp $job_to_send ${job_catalog}/$job_on_remhost
	931	fi
	932
	933
	934
	935	# NQS- BZW. LOADLEVELER-JOB STARTEN
	936	if [[ $remote_host != $local_host ]]
	937	then
	938	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[82]	939	if [[ $remote_host != lctit ]]
[1]	940	then
[82]	941	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[1]	942	else
[82]	943	# TIT ERLAUBT NUR DIE AUSFÜHRUNG GANZ BESTIMMTER KOMMANDOS
	944	# MIT SSH, DESHALB AUFRUF PER PIPE
	945	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
	946	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	947	echo "first try" > send_protocol
	948	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
	949	do
	950	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
	951	then
	952	printf "\n trying to submit job to TIT cluster..."
	953	else
	954	printf "\n +++ failed ... trying to submit again ..."
	955	fi
	956	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
	957	cat send_protocol
	958	done
	959	sleep 10
	960	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	961	echo "$submcom $job_on_remhost"
	962	rm send_protocol
	963	fi
[1]	964
	965	[[ $verify = true ]] && printf " >>> o.k.\n"
	966	else
	967	cd $job_catalog
	968	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
	969	then
	970	eval $submcom $job_on_remhost
	971	elif [[ $local_host = lctit ]]
	972	then
	973	eval $submcom $job_on_remhost
	974	echo "$submcom $job_on_remhost"
	975	chmod u+x $job_on_remhost
	976	elif [[ $local_host = nech ]]
	977	then
	978	if [[ $queue = default ]]
	979	then
	980	eval $submcom $job_on_remhost
	981	else
	982	eval $submcom -q $queue $job_on_remhost
	983	fi
	984	else
	985	qsub $job_on_remhost
	986	fi
[69]	987	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
	988	[[ $local_host != lctit ]] && rm $job_on_remhost
[1]	989	cd - > /dev/null
	990	fi
	991	fi
	992
	993
	994
	995	# ABSCHLUSSARBEITEN
	996
	997	if [[ $no_submit = false ]]
	998	then
	999	rm $job_to_send
	1000	fi
	1001	[[ $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 |