Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 114

Last change on this file since 114 was 113, checked in by raasch, 17 years ago
job classes csoloh... added for hanni
Property svn:keywords set to `Rev`
File size: 36.5 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...)
[1]	87
	88
[113]	89
[1]	90	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
	91	delete_dayfile=false
	92	locat=normal
	93	no_default_queue=none
	94	no_submit=false
	95	job_catalog="~/job_queue"
	96	job_name=none
	97	local_user=$LOGNAME
	98	node_usage=shared
	99	numprocs=1
	100	punkte="..........................................................."
	101	submcom=qsub
	102	queue=default
	103	remote_host=none
	104	remote_user=""
	105	verify=true
	106
	107	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
	108	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
	109	typeset -L20 spalte1
	110	typeset -R40 spalte2
	111	typeset -L60 spalte3
	112
	113
	114
	115	# FEHLERBEHANDLUNG
	116	# BEI EXIT:
	117	trap 'if [[ $locat != normal ]]
	118	then
	119	case $locat in
	120	(option) printf "\n --> available optios can be displayed"
	121	printf " by typing:"
	122	printf "\n \"subjob ?\" \n";;
	123	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	124	(*) printf "\n +++ unknown error"
	125	printf "\n please inform S. Raasch!\n"
	126	esac
	127	[[ -f $job_to_send ]] && rm $job_to_send
	128	printf "\n\n+++ SUBJOB killed \n\n"
	129	fi' exit
	130
	131
	132	# BEI TERMINAL-BREAK:
	133	trap '[[ -f $job_to_send ]] && rm $job_to_send
	134	printf "\n\n+++ SUBJOB killed \n\n"
	135	exit
	136	' 2
	137
	138
	139
	140
	141	# LOKALEN HOSTNAMEN FESTSTELLEN
	142	local_host=$(hostname)
	143
	144
	145
	146	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
	147	# UEBERHAUPT ZULAESSIG IST
	148	case $local_host in
	149	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[102]	150	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	151	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[1]	152	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
[12]	153	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[108]	154	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
[1]	155	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	156	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[108]	157	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
[101]	158	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[35]	159	(gate\|n-sx) local_addres=133.5.178.11; local_host=neck;;
[1]	160	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	161	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	162	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	163	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[1]	164	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
[5]	165	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
	166	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
	167	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[1]	168	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	169	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	170	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
	171	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[1]	172	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
	173	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
	174	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
	175	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
	176	(*) printf "\n +++ \"$local_host\" unknown";
	177	printf "\n please inform S. Raasch!";
	178	locat=parameter; exit;;
	179	esac
	180
	181
	182
	183	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	184	remote_host=$local_host
	185
	186
	187
	188
	189	# PROZEDUROPTIONEN EINLESEN
[82]	190	while getopts :c:dDh:m:n:N:O:q:t:T:u:vX: option
[1]	191	do
	192	case $option in
	193	(c) job_catalog=$OPTARG;;
	194	(d) delete_dayfile=true;;
	195	(D) no_submit=true;;
	196	(h) remote_host=$OPTARG;;
	197	(m) memory=$OPTARG;;
	198	(n) job_name=$OPTARG;;
	199	(N) node_usage=$OPTARG;;
	200	(O) threads_per_task=$OPTARG;;
	201	(q) no_default_queue=$OPTARG;;
	202	(t) cputime=$OPTARG;;
	203	(T) tasks_per_node=$OPTARG;;
	204	(u) remote_user=$OPTARG;;
	205	(v) verify=false;;
	206	(X) numprocs=$OPTARG;;
	207	(\?) printf "\n +++ Option $OPTARG unknown \n";
	208	locat=option; exit;;
	209	esac
	210	done
	211
	212
	213	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	214	shift OPTIND-1; file_to_send=$1
	215
	216
	217	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	218	if [ "$1" = "?" ]
	219	then
	220	(printf "\n *** subjob can be called as follows:\n"
	221	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	222	printf "\n Description of available options:\n"
	223	printf "\n Option Description Default-Value"
	224	printf "\n -c job-input- and output-catalog ~/job_queue"
	225	printf "\n -d no job-protocol will be created ---"
	226	printf "\n -D only the job-file will be created ---"
	227	printf "\n -h execution host, available hosts: $remote_host"
[5]	228	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
[35]	229	printf "\n lctit, nech, neck, unics"
[1]	230	printf "\n -m memory demand per process in MByte ---"
	231	printf "\n -n jobname <jobdatei>"
	232	printf "\n -O threads per task (for OpenMP usage) 1"
	233	printf "\n -q job-queue to be used default"
	234	printf "\n -t allowed cpu-time in seconds ---"
	235	printf "\n -T tasks per node (on parallel hosts) ---"
	236	printf "\n -u username on execution host from .netrc"
	237	printf "\n -v no prompt for confirmation ---"
	238	printf "\n -X # of processors (on parallel hosts) 1"
	239	printf "\n "
	240	printf "\n The only possible positional parameter is <jobfile>:"
	241	printf "\n The complete NQS-job must be provided here."
	242	printf "\n <jobfile>=? creates this outline\n\n") \| more
	243	exit
	244	fi
	245
	246
	247
	248	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	249	if [[ "$file_to_send" = "" ]]
	250	then
	251	printf "\n +++ job-file missing"
	252	locat=parameter; exit
	253	else
	254	if [[ -f $file_to_send ]]
	255	then
	256	true
	257	else
	258	printf "\n +++ job-file: "
	259	printf "\n $file_to_send"
	260	printf "\n does not exist"
	261	locat=parameter; exit
	262	fi
	263	fi
	264
	265
	266
	267	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	268	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	269	if [[ $job_name = none ]]
	270	then
	271	job_name=$file_to_send
	272	fi
	273	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	274	then
	275	printf "\n +++ job-file name: "
	276	printf "\n $job_name"
	277	printf "\n must not contain \"/\"-characters"
	278	locat=parameter; exit
	279	fi
	280
	281
	282
	283
	284	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	285	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	286	if [[ $remote_host = none ]]
	287	then
	288	printf "\n +++ host missing"
	289	locat=option; exit
	290	else
	291	case $remote_host in
	292	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	293	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	294	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	295	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	296	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[108]	297	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
[69]	298	(lctit) queue=default; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
[1]	299	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
[35]	300	(neck) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=133.5.178.11; submcom="/usr/bin/nqsII/qsub";;
[1]	301	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	302	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	303	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	304	locat=parameter; exit;;
	305	esac
	306	fi
	307
	308
	309	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	310	if [[ $no_default_queue != none ]]
	311	then
	312	error=false
	313	ndq=$no_default_queue
	314	case $remote_host in
	315	(ibm) case $ndq in
	316	(p690_express\|p690_standard\|p690_long) error=false;;
	317	(*) error=true;;
	318	esac;;
	319	(ibmb) case $ndq in
	320	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
	321	(*) error=true;;
	322	esac;;
	323	(ibmh) case $ndq in
[113]	324	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|csoloh3h\|csoloh6h\|csoloh12h\|csoloh1d\|csoloh2d\|cspec) error=false;;
[1]	325	(*) error=true;;
	326	esac;;
	327	(ibms) case $ndq in
	328	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	329	(*) error=true;;
	330	esac;;
	331	(ibmy) case $ndq in
	332	(parallel) error=false;;
	333	(*) error=true;;
	334	esac;;
	335	(lctit) case $ndq in
[69]	336	(default\|high\|bes1\|bes2\|sla1\|sla2) error=false;;
[1]	337	(*) error=true;;
	338	esac;;
	339	(t3eb) case $ndq in
	340	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	341	(*) error=true;;
	342	esac;;
	343	(t3eh) case $ndq in
	344	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	345	(*) error=true;;
	346	esac;;
	347	(t3ej2\|t3ej5) case $ndq in
	348	(low\|normal\|high) error=false;;
	349	(*) error=true;;
	350	esac;;
	351	(t3es) case $ndq in
	352	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	353	(*) error=true;;
	354	esac;;
	355	(unics) case $ndq in
	356	(unics\|ht) error=false;;
	357	(*) error=true;;
	358	esac;;
	359	esac
	360	if [[ $error = true ]]
	361	then
	362	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	363	locat=parameter; exit
	364	else
	365	queue=$no_default_queue
	366	fi
	367	fi
	368
	369
	370
	371	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
	372	if [[ $node_usage != shared && $queue = cdev ]]
	373	then
	374	node_usage=shared
	375	fi
	376
	377
	378
	379	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	380	# AUFTEILEN
	381	done=false
	382	while [[ $done = false ]]
	383	do
	384	if (( $cputime <= 0 ))
	385	then
	386	printf "\n +++ wrong cpu-time or cpu-time missing"
	387	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	388	printf "\n >>> "
	389	read cputime 1>/dev/null 2>&1
	390	else
	391	done=true
	392	fi
	393	done
	394	if [[ $remote_host = nech ]]
	395	then
	396	if (( tasks_per_node != 0 ))
	397	then
	398	(( cputime = cputime * tasks_per_node ))
	399	elif [[ $numprocs != 0 ]]
	400	then
	401	(( cputime = cputime * numprocs ))
	402	fi
	403	fi
	404	(( stunden = cputime / 3600 ))
	405	(( resttime = cputime - stunden * 3600 ))
	406	(( minuten = resttime / 60 ))
	407	(( sekunden = resttime - minuten * 60 ))
	408	timestring=${stunden}:${minuten}:${sekunden}
	409
	410
	411
	412	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	413	done=false
	414	while [[ $done = false ]]
	415	do
	416	if (( memory <= 0 ))
	417	then
	418	printf "\n +++ wrong memory demand or memory demand missing"
	419	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	420	printf "\n >>> "
	421	read memory 1>/dev/null 2>&1
	422	else
	423	done=true
	424	fi
	425	done
	426
[35]	427	if [[ $remote_host = nech \|\| $remote_host = neck ]]
[1]	428	then
	429	if (( tasks_per_node != 0 ))
	430	then
	431	(( Memory = memory * tasks_per_node / 1000 ))
	432	elif [[ $numprocs != 0 ]]
	433	then
	434	(( Memory = memory * numprocs / 1000 ))
	435	else
	436	(( Memory = memory / 1000 ))
	437	fi
	438	fi
	439
	440
	441	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	442	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	443	then
	444	(( memory = memory * threads_per_task ))
	445	fi
	446
	447
	448	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	449	if (( tasks_per_node != 0 ))
	450	then
	451	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
	452	fi
	453
	454
	455
	456	# HEADER-AUSGABE
	457	if [[ $verify = true ]]
	458	then
	459	printf "\n\n"
	460	printf "#--------------------------------------------------------------# \n"
	461	spalte1=SUBJOB;spalte2=$(date)
	462	printf "\| $spalte1$spalte2 \| \n"
	463	printf "\| \| \n"
	464	printf "\| values of parameters/options: \| \n"
	465	spalte1=local_host$punkte; spalte2=$punkte$local_host
	466	printf "\| $spalte1$spalte2 \| \n"
	467	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	468	printf "\| $spalte1$spalte2 \| \n"
	469	spalte1=queue$punkte; spalte2=$punkte$queue
	470	printf "\| $spalte1$spalte2 \| \n"
	471	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	472	printf "\| $spalte1$spalte2 \| \n"
	473	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	474	printf "\| $spalte1$spalte2 \| \n"
	475	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	476	printf "\| $spalte1$spalte2 \| \n"
	477	printf "#--------------------------------------------------------------# \n\n"
	478
	479
	480	# KONTROLLABFRAGE, OB ALLES O.K.
	481	antwort="dummy"
	482	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	483	do
	484	read antwort?" >>> continue (y/n) ? "
	485	done
	486	if [[ $antwort = n \|\| $antwort = N ]]
	487	then
	488	locat=verify; exit
	489	fi
	490	printf "\n"
	491	fi
	492
	493
	494
	495	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	496	kennung=$RANDOM
	497	job_on_remhost=${job_name}_${kennung}_$local_host
	498	job_to_send=job_to_send_$kennung
	499	if [[ $delete_dayfile = false ]]
	500	then
	501	remote_dayfile=${local_host}_${job_name}_result_$kennung
	502	local_dayfile=${remote_host}_${job_name}
	503	else
	504	remote_dayfile=/dev/null
	505	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	506	then
	507	notification='# @ notification = never'
	508	fi
	509	fi
	510
	511
	512
	513	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
	514	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	515	then
	516
	517	if [[ $remote_host = ibmy ]]
	518	then
	519	consumable_memory=""
	520	else
	521	consumable_memory="ConsumableMemory($memory mb)"
	522	fi
	523	if [[ $queue = cdev ]]
	524	then
	525	data_limit="# @ data_limit = 1.76gb"
	526	network_to_use="# @ network.mpi = sn_all,shared,ip"
	527	else
	528	if [[ $remote_host = ibms ]]
	529	then
	530	network_to_use="# @ network.mpi = csss,shared,us"
	531	elif [[ $remote_host = ibmy ]]
	532	then
	533	network_to_use=""
	534	else
	535	network_to_use="# @ network.mpi = sn_all,shared,us"
	536	data_limit="# @ data_limit = 1.76gb"
	537	fi
	538	fi
	539
	540	cat > $job_to_send << %%END%%
	541	#!/bin/ksh
	542
	543	# @ job_type = parallel
	544	# @ job_name = palm
	545	# @ wall_clock_limit = ${timestring},$timestring
	546	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	547	# @ output = $remote_dayfile
	548	# @ error = $remote_dayfile
	549	# @ image_size = 50
	550	# @ class = $queue
	551	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
	552	$network_to_use
	553	$data_limit
	554	$notification
	555
	556	%%END%%
	557
	558	if (( nodes > 0 ))
	559	then
	560
	561	cat >> $job_to_send << %%END%%
	562	# @ node = $nodes
	563	# @ tasks_per_node = $tasks_per_node
	564	# @ node_usage = $node_usage
	565	# @ queue
	566
	567	%%END%%
	568
	569	else
	570
	571	if [[ $remote_host != ibmy ]]
	572	then
	573
	574	cat >> $job_to_send << %%END%%
	575	# @ blocking = unlimited
	576	# @ total_tasks = $numprocs
	577	# @ node_usage = $node_usage
	578	# @ queue
	579
	580	%%END%%
	581
	582	else
	583
	584	cat >> $job_to_send << %%END%%
	585	# @ node = 1
	586	# @ total_tasks = $numprocs
	587	# @ queue
	588
	589	%%END%%
	590
	591	fi
	592
	593	fi
	594
	595	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	596	then
	597
	598	cat > $job_to_send << %%END%%
	599	#!/bin/ksh
	600
	601	# @ job_type = serial
	602	# @ node_usage = $node_usage
	603	# @ job_name = palm
	604	# @ wall_clock_limit = ${timestring},$timestring
	605	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	606	# @ output = $remote_dayfile
	607	# @ error = $remote_dayfile
	608	# @ class = $queue
	609	$notification
	610
	611	# @ queue
	612
	613	%%END%%
	614
[108]	615	elif [[ $remote_host = lcfimm ]]
	616	then
	617
	618	if [[ $numprocs != 0 ]]
	619	then
	620	cat > $job_to_send << %%END%%
	621	#!/bin/ksh
	622	#PBS -N $job_name
	623	#PBS -A nersc
	624	#PBS -l walltime=$timestring
	625	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
	626	#PBS -l pmem=${memory}mb
	627	#PBS -m abe
	628	#PBS -M igore@nersc.no
	629	#PBS -o $remote_dayfile
	630	#PBS -j oe
	631	mpd &
	632
	633	%%END%%
	634
	635	else
	636	cat > $job_to_send << %%END%%
	637	#!/bin/ksh
	638	#PBS -N $job_name
	639	#PBS -A nersc
	640	#PBS -l walltime=$timestring
	641	#PBS -l ncpus=1
	642	#PBS -l pmem=${memory}mb
	643	#PBS -m abe
	644	#PBS -M igore@nersc.no
	645	#PBS -o $remote_dayfile
	646	#PBS -j oe
	647
	648	%%END%%
	649
	650	fi
	651
[1]	652	elif [[ $remote_host = nech ]]
	653	then
	654
	655	if (( nodes > 1 ))
	656	then
	657	cat > $job_to_send << %%END%%
	658	#!/bin/ksh
	659	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	660	#PBS -l ${qsubmem}=${Memory}gb
	661	#PBS -b $nodes
	662	#PBS -o $remote_dayfile
	663	#PBS -N palm
	664	#PBS -j o
	665	#PBS -T mpisx
	666
	667	%%END%%
	668
	669	elif [[ $numprocs != 0 ]]
	670	then
	671	cat > $job_to_send << %%END%%
	672	#!/bin/ksh
	673	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	674	#PBS -l ${qsubmem}=${Memory}gb
	675	#PBS -o $remote_dayfile
	676	#PBS -N palm
	677	#PBS -j o
	678
	679	%%END%%
	680
	681	else
	682	cat > $job_to_send << %%END%%
	683	#!/bin/ksh
	684	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	685	#PBS -o $remote_dayfile
	686	#PBS -j o
	687
	688	%%END%%
	689
	690	fi
	691
[35]	692	elif [[ $remote_host = neck ]]
	693	then
	694
	695	if (( nodes > 1 ))
	696	then
	697	cat > $job_to_send << %%END%%
	698	#!/bin/ksh
	699	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	700	#PBS -l ${qsubmem}=${Memory}gb
	701	#PBS -b $nodes
	702	#PBS -o $remote_dayfile
	703	#PBS -N palm
	704	#PBS -j o
	705	#PBS -T mpisx
	706
	707	%%END%%
	708
	709	elif [[ $numprocs != 0 ]]
	710	then
	711	cat > $job_to_send << %%END%%
	712	#!/bin/ksh
	713	#PBS -q ${queue}
	714	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	715	#PBS -l ${qsubmem}=${Memory}gb
	716	#PBS -o $remote_dayfile
	717	#PBS -N palm
	718	#PBS -j o
	719	#PBS -T mpisx
	720
	721	%%END%%
	722
	723	else
	724	cat > $job_to_send << %%END%%
	725	#!/bin/ksh
	726	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	727	#PBS -o $remote_dayfile
	728	#PBS -j o
	729
	730	%%END%%
	731
	732	fi
	733
[1]	734	elif [[ $remote_host = lctit ]]
	735	then
	736	cat > $job_to_send << %%END%%
	737	#!/bin/bash
	738	#$ -S /bin/bash
	739	cd $job_catalog
[69]	740	export PATH=$PALM_BIN:\$PATH
	741	echo \$PATH
[1]	742
	743	%%END%%
	744
	745	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
	746	submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
	747
	748	else
	749
	750	cat > $job_to_send << %%END%%
	751	# @\$-q ${queue}
	752	# @\$-l${qsubtime} $timestring
	753	# @\$-l${qsubmem} ${memory}mb
	754	# @\$-o $remote_dayfile
	755	# @\$-eo
	756
	757	%%END%%
	758
	759	fi
	760
	761
	762	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	763	# BEI EXIT VERANLASSEN
	764	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	765	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	766	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	767	then
	768	echo "set +vx" >> $job_to_send
	769	echo "trap '" >> $job_to_send
	770	echo "set +vx" >> $job_to_send
[35]	771	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
[1]	772	then
	773	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
	774	then
	775	return_queue=c1
	776	elif [[ $remote_host = ibms ]]
	777	then
	778	return_queue=p_normal
	779	elif [[ $remote_host = ibmy ]]
	780	then
	781	return_queue=serial
	782	elif [[ $remote_host = lctit ]]
	783	then
	784	return_queue=default
[35]	785	elif [[ $remote_host = neck ]]
[1]	786	then
[35]	787	return_queue=S
[1]	788	else
	789	return_queue=unknown
	790	fi
	791
	792	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	793	then
	794
	795	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	796	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	797	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	798	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	799	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	800	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	801	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	802	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	803	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	804	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	805
	806	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	807	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	808
	809	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	810	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
	811	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	812
	813	elif [[ $remote_host = nech ]]
	814	then
	815	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	816	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	817	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	818	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	819	echo "#PBS -j o" >> $job_to_send
	820	echo " " >> $job_to_send
	821	echo "set -x" >> $job_to_send
	822	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	823	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	824	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	825	echo "%%END%%" >> $job_to_send
	826
[35]	827	elif [[ $remote_host = neck ]]
	828	then
	829	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	830	echo "#PBS -q $return_queue" >> $job_to_send
	831	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	832	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	833	echo "#PBS -j o" >> $job_to_send
	834	echo " " >> $job_to_send
	835	echo "set -x" >> $job_to_send
	836	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	837	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	838	echo "%%END%%" >> $job_to_send
	839
[1]	840	elif [[ $remote_host = lctit ]]
	841	then
	842	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	843	echo "set -x" >> $job_to_send
	844	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
	845	echo "cd $job_catalog" >> $job_to_send
	846	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	847	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	848	echo "%%END%%" >> $job_to_send
	849
	850	else
	851
	852	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	853	echo "# @\\\$-q $return_queue" >> $job_to_send
	854	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	855	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	856	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	857	then
	858	echo "# @\$-l mpp_p=0" >> $job_to_send
	859	fi
	860	echo '# @\$-lF 10mb' >> $job_to_send
	861	# echo '# @\$-o /dev/null' >> $job_to_send
	862	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	863	echo '# @\\\$-eo' >> $job_to_send
	864	echo " " >> $job_to_send
	865	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	866	then
	867	echo "set +vx" >> $job_to_send
	868	echo ". .profile" >> $job_to_send
	869	fi
	870	echo "set -x" >> $job_to_send
	871	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	872	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	873	echo "%%END%%" >> $job_to_send
	874	fi
	875
	876	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	877	then
	878	echo "llsubmit scpjob.$kennung" >> $job_to_send
	879	elif [[ $remote_host = lctit ]]
	880	then
	881	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	882	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
	883	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
	884	echo "sleep 10" >> $job_to_send
	885	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	886	then
	887	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	888	elif [[ $remote_host = t3es ]]
	889	then
	890	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
	891	else
	892	echo "qsub scpjob.$kennung" >> $job_to_send
	893	fi
	894	echo "rm scpjob.$kennung" >> $job_to_send
	895	if [[ $remote_host = nech ]]
	896	then
	897	echo "cd -" >> $job_to_send
	898	fi
	899	else
	900	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	901	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	902	fi
	903	echo "set -x" >> $job_to_send
	904	echo " ' exit" >> $job_to_send
	905	echo "set -x" >> $job_to_send
	906	fi
	907
	908
	909
	910	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	911	cat $file_to_send >> $job_to_send
[69]	912
[1]	913	if [[ $remote_host = ibm ]]
	914	then
	915	echo " " >> $job_to_send
	916	echo "exit" >> $job_to_send
	917	fi
[69]	918	if [[ $remote_host = lctit ]]
	919	then
	920	echo " " >> $job_to_send
	921	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	922	fi
[1]	923
	924
	925
	926	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	927	if [[ -z $remote_user ]]
	928	then
	929	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	930	then
	931	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	932	fi
	933	fi
	934
	935
	936
	937	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	938	if [[ $no_submit = false ]]
	939	then
	940	if [[ $remote_host != $local_host ]]
	941	then
	942	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	943	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	944	then
[82]	945	job_catalog_save=$job_catalog
	946	job_catalog=job_queue
	947	elif [[ $remote_host = nech ]]
	948	then
	949	job_catalog_save=$job_catalog
	950	job_catalog=/hpf/b/${remote_user}/job_queue
	951	fi
	952	if [[ $local_host = decalpha ]]
	953	then
	954	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	955	# PASSWORT AB
	956	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	957	elif [[ $remote_host = nech ]]
	958	then
	959	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	960	# TRANSFERIERT WERDEN
	961	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	962	else
[82]	963	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	964	fi
[82]	965	if [[ $? = 1 ]]
	966	then
	967	locat=scp; exit
	968	fi
	969	if [[ $remote_host = ibms ]]
	970	then
	971	job_catalog=$job_catalog_save
	972	fi
[1]	973	[[ $verify = true ]] && printf "\n >>> finished\n"
	974	else
	975	eval job_catalog=$job_catalog
	976	cp $job_to_send ${job_catalog}/$job_on_remhost
	977	fi
	978
	979
	980
	981	# NQS- BZW. LOADLEVELER-JOB STARTEN
	982	if [[ $remote_host != $local_host ]]
	983	then
	984	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[82]	985	if [[ $remote_host != lctit ]]
[1]	986	then
[82]	987	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[1]	988	else
[82]	989	# TIT ERLAUBT NUR DIE AUSFÜHRUNG GANZ BESTIMMTER KOMMANDOS
	990	# MIT SSH, DESHALB AUFRUF PER PIPE
	991	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
	992	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	993	echo "first try" > send_protocol
	994	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
	995	do
	996	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
	997	then
	998	printf "\n trying to submit job to TIT cluster..."
	999	else
	1000	printf "\n +++ failed ... trying to submit again ..."
	1001	fi
	1002	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
	1003	cat send_protocol
	1004	done
	1005	sleep 10
	1006	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	1007	echo "$submcom $job_on_remhost"
	1008	rm send_protocol
	1009	fi
[1]	1010
	1011	[[ $verify = true ]] && printf " >>> o.k.\n"
	1012	else
	1013	cd $job_catalog
	1014	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
	1015	then
	1016	eval $submcom $job_on_remhost
[108]	1017	elif [[ $local_host = lcfimm ]]
	1018	then
	1019	eval $submcom $job_on_remhost
	1020	echo "$submcom $job_on_remhost"
	1021	chmod u+x $job_on_remhost
[1]	1022	elif [[ $local_host = lctit ]]
	1023	then
	1024	eval $submcom $job_on_remhost
	1025	echo "$submcom $job_on_remhost"
	1026	chmod u+x $job_on_remhost
	1027	elif [[ $local_host = nech ]]
	1028	then
	1029	if [[ $queue = default ]]
	1030	then
	1031	eval $submcom $job_on_remhost
	1032	else
	1033	eval $submcom -q $queue $job_on_remhost
	1034	fi
	1035	else
	1036	qsub $job_on_remhost
	1037	fi
[69]	1038	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
	1039	[[ $local_host != lctit ]] && rm $job_on_remhost
[1]	1040	cd - > /dev/null
	1041	fi
	1042	fi
	1043
	1044
	1045
	1046	# ABSCHLUSSARBEITEN
	1047
	1048	if [[ $no_submit = false ]]
	1049	then
	1050	rm $job_to_send
	1051	fi
	1052	[[ $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 |