Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 17

Last change on this file since 17 was 12, checked in by raasch, 18 years ago
breva admitted for subjob
File size: 36.0 KB

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