Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 101

Last change on this file since 101 was 101, checked in by raasch, 17 years ago
small changes in documentation and subjob
Property svn:keywords set to `Rev`
File size: 35.3 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 $local_host in
	145	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[5]	146	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[1]	147	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
[12]	148	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[1]	149	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	150	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[101]	151	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[35]	152	(gate\|n-sx) local_addres=133.5.178.11; local_host=neck;;
[1]	153	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	154	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	155	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	156	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[1]	157	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
[5]	158	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
	159	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
	160	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[1]	161	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	162	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	163	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
	164	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[1]	165	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
	166	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
	167	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
	168	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
	169	(*) printf "\n +++ \"$local_host\" unknown";
	170	printf "\n please inform S. Raasch!";
	171	locat=parameter; exit;;
	172	esac
	173
	174
	175
	176	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	177	remote_host=$local_host
	178
	179
	180
	181
	182	# PROZEDUROPTIONEN EINLESEN
[82]	183	while getopts :c:dDh:m:n:N:O:q:t:T:u:vX: option
[1]	184	do
	185	case $option in
	186	(c) job_catalog=$OPTARG;;
	187	(d) delete_dayfile=true;;
	188	(D) no_submit=true;;
	189	(h) remote_host=$OPTARG;;
	190	(m) memory=$OPTARG;;
	191	(n) job_name=$OPTARG;;
	192	(N) node_usage=$OPTARG;;
	193	(O) threads_per_task=$OPTARG;;
	194	(q) no_default_queue=$OPTARG;;
	195	(t) cputime=$OPTARG;;
	196	(T) tasks_per_node=$OPTARG;;
	197	(u) remote_user=$OPTARG;;
	198	(v) verify=false;;
	199	(X) numprocs=$OPTARG;;
	200	(\?) printf "\n +++ Option $OPTARG unknown \n";
	201	locat=option; exit;;
	202	esac
	203	done
	204
	205
	206	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	207	shift OPTIND-1; file_to_send=$1
	208
	209
	210	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	211	if [ "$1" = "?" ]
	212	then
	213	(printf "\n *** subjob can be called as follows:\n"
	214	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	215	printf "\n Description of available options:\n"
	216	printf "\n Option Description Default-Value"
	217	printf "\n -c job-input- and output-catalog ~/job_queue"
	218	printf "\n -d no job-protocol will be created ---"
	219	printf "\n -D only the job-file will be created ---"
	220	printf "\n -h execution host, available hosts: $remote_host"
[5]	221	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
[35]	222	printf "\n lctit, nech, neck, unics"
[1]	223	printf "\n -m memory demand per process in MByte ---"
	224	printf "\n -n jobname <jobdatei>"
	225	printf "\n -O threads per task (for OpenMP usage) 1"
	226	printf "\n -q job-queue to be used default"
	227	printf "\n -t allowed cpu-time in seconds ---"
	228	printf "\n -T tasks per node (on parallel hosts) ---"
	229	printf "\n -u username on execution host from .netrc"
	230	printf "\n -v no prompt for confirmation ---"
	231	printf "\n -X # of processors (on parallel hosts) 1"
	232	printf "\n "
	233	printf "\n The only possible positional parameter is <jobfile>:"
	234	printf "\n The complete NQS-job must be provided here."
	235	printf "\n <jobfile>=? creates this outline\n\n") \| more
	236	exit
	237	fi
	238
	239
	240
	241	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	242	if [[ "$file_to_send" = "" ]]
	243	then
	244	printf "\n +++ job-file missing"
	245	locat=parameter; exit
	246	else
	247	if [[ -f $file_to_send ]]
	248	then
	249	true
	250	else
	251	printf "\n +++ job-file: "
	252	printf "\n $file_to_send"
	253	printf "\n does not exist"
	254	locat=parameter; exit
	255	fi
	256	fi
	257
	258
	259
	260	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	261	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	262	if [[ $job_name = none ]]
	263	then
	264	job_name=$file_to_send
	265	fi
	266	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	267	then
	268	printf "\n +++ job-file name: "
	269	printf "\n $job_name"
	270	printf "\n must not contain \"/\"-characters"
	271	locat=parameter; exit
	272	fi
	273
	274
	275
	276
	277	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	278	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	279	if [[ $remote_host = none ]]
	280	then
	281	printf "\n +++ host missing"
	282	locat=option; exit
	283	else
	284	case $remote_host in
	285	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	286	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	287	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	288	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	289	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[69]	290	(lctit) queue=default; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
[1]	291	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
[35]	292	(neck) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=133.5.178.11; submcom="/usr/bin/nqsII/qsub";;
[1]	293	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	294	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	295	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	296	locat=parameter; exit;;
	297	esac
	298	fi
	299
	300
	301	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	302	if [[ $no_default_queue != none ]]
	303	then
	304	error=false
	305	ndq=$no_default_queue
	306	case $remote_host in
	307	(ibm) case $ndq in
	308	(p690_express\|p690_standard\|p690_long) error=false;;
	309	(*) error=true;;
	310	esac;;
	311	(ibmb) case $ndq in
	312	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
	313	(*) error=true;;
	314	esac;;
	315	(ibmh) case $ndq in
	316	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|cspec) error=false;;
	317	(*) error=true;;
	318	esac;;
	319	(ibms) case $ndq in
	320	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	321	(*) error=true;;
	322	esac;;
	323	(ibmy) case $ndq in
	324	(parallel) error=false;;
	325	(*) error=true;;
	326	esac;;
	327	(lctit) case $ndq in
[69]	328	(default\|high\|bes1\|bes2\|sla1\|sla2) error=false;;
[1]	329	(*) error=true;;
	330	esac;;
	331	(t3eb) case $ndq in
	332	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	333	(*) error=true;;
	334	esac;;
	335	(t3eh) case $ndq in
	336	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	337	(*) error=true;;
	338	esac;;
	339	(t3ej2\|t3ej5) case $ndq in
	340	(low\|normal\|high) error=false;;
	341	(*) error=true;;
	342	esac;;
	343	(t3es) case $ndq in
	344	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	345	(*) error=true;;
	346	esac;;
	347	(unics) case $ndq in
	348	(unics\|ht) error=false;;
	349	(*) error=true;;
	350	esac;;
	351	esac
	352	if [[ $error = true ]]
	353	then
	354	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	355	locat=parameter; exit
	356	else
	357	queue=$no_default_queue
	358	fi
	359	fi
	360
	361
	362
	363	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
	364	if [[ $node_usage != shared && $queue = cdev ]]
	365	then
	366	node_usage=shared
	367	fi
	368
	369
	370
	371	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	372	# AUFTEILEN
	373	done=false
	374	while [[ $done = false ]]
	375	do
	376	if (( $cputime <= 0 ))
	377	then
	378	printf "\n +++ wrong cpu-time or cpu-time missing"
	379	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	380	printf "\n >>> "
	381	read cputime 1>/dev/null 2>&1
	382	else
	383	done=true
	384	fi
	385	done
	386	if [[ $remote_host = nech ]]
	387	then
	388	if (( tasks_per_node != 0 ))
	389	then
	390	(( cputime = cputime * tasks_per_node ))
	391	elif [[ $numprocs != 0 ]]
	392	then
	393	(( cputime = cputime * numprocs ))
	394	fi
	395	fi
	396	(( stunden = cputime / 3600 ))
	397	(( resttime = cputime - stunden * 3600 ))
	398	(( minuten = resttime / 60 ))
	399	(( sekunden = resttime - minuten * 60 ))
	400	timestring=${stunden}:${minuten}:${sekunden}
	401
	402
	403
	404	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	405	done=false
	406	while [[ $done = false ]]
	407	do
	408	if (( memory <= 0 ))
	409	then
	410	printf "\n +++ wrong memory demand or memory demand missing"
	411	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	412	printf "\n >>> "
	413	read memory 1>/dev/null 2>&1
	414	else
	415	done=true
	416	fi
	417	done
	418
[35]	419	if [[ $remote_host = nech \|\| $remote_host = neck ]]
[1]	420	then
	421	if (( tasks_per_node != 0 ))
	422	then
	423	(( Memory = memory * tasks_per_node / 1000 ))
	424	elif [[ $numprocs != 0 ]]
	425	then
	426	(( Memory = memory * numprocs / 1000 ))
	427	else
	428	(( Memory = memory / 1000 ))
	429	fi
	430	fi
	431
	432
	433	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	434	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	435	then
	436	(( memory = memory * threads_per_task ))
	437	fi
	438
	439
	440	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	441	if (( tasks_per_node != 0 ))
	442	then
	443	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
	444	fi
	445
	446
	447
	448	# HEADER-AUSGABE
	449	if [[ $verify = true ]]
	450	then
	451	printf "\n\n"
	452	printf "#--------------------------------------------------------------# \n"
	453	spalte1=SUBJOB;spalte2=$(date)
	454	printf "\| $spalte1$spalte2 \| \n"
	455	printf "\| \| \n"
	456	printf "\| values of parameters/options: \| \n"
	457	spalte1=local_host$punkte; spalte2=$punkte$local_host
	458	printf "\| $spalte1$spalte2 \| \n"
	459	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	460	printf "\| $spalte1$spalte2 \| \n"
	461	spalte1=queue$punkte; spalte2=$punkte$queue
	462	printf "\| $spalte1$spalte2 \| \n"
	463	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	464	printf "\| $spalte1$spalte2 \| \n"
	465	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	466	printf "\| $spalte1$spalte2 \| \n"
	467	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	468	printf "\| $spalte1$spalte2 \| \n"
	469	printf "#--------------------------------------------------------------# \n\n"
	470
	471
	472	# KONTROLLABFRAGE, OB ALLES O.K.
	473	antwort="dummy"
	474	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	475	do
	476	read antwort?" >>> continue (y/n) ? "
	477	done
	478	if [[ $antwort = n \|\| $antwort = N ]]
	479	then
	480	locat=verify; exit
	481	fi
	482	printf "\n"
	483	fi
	484
	485
	486
	487	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	488	kennung=$RANDOM
	489	job_on_remhost=${job_name}_${kennung}_$local_host
	490	job_to_send=job_to_send_$kennung
	491	if [[ $delete_dayfile = false ]]
	492	then
	493	remote_dayfile=${local_host}_${job_name}_result_$kennung
	494	local_dayfile=${remote_host}_${job_name}
	495	else
	496	remote_dayfile=/dev/null
	497	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	498	then
	499	notification='# @ notification = never'
	500	fi
	501	fi
	502
	503
	504
	505	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
	506	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	507	then
	508
	509	if [[ $remote_host = ibmy ]]
	510	then
	511	consumable_memory=""
	512	else
	513	consumable_memory="ConsumableMemory($memory mb)"
	514	fi
	515	if [[ $queue = cdev ]]
	516	then
	517	data_limit="# @ data_limit = 1.76gb"
	518	network_to_use="# @ network.mpi = sn_all,shared,ip"
	519	else
	520	if [[ $remote_host = ibms ]]
	521	then
	522	network_to_use="# @ network.mpi = csss,shared,us"
	523	elif [[ $remote_host = ibmy ]]
	524	then
	525	network_to_use=""
	526	else
	527	network_to_use="# @ network.mpi = sn_all,shared,us"
	528	data_limit="# @ data_limit = 1.76gb"
	529	fi
	530	fi
	531
	532	cat > $job_to_send << %%END%%
	533	#!/bin/ksh
	534
	535	# @ job_type = parallel
	536	# @ job_name = palm
	537	# @ wall_clock_limit = ${timestring},$timestring
	538	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	539	# @ output = $remote_dayfile
	540	# @ error = $remote_dayfile
	541	# @ image_size = 50
	542	# @ class = $queue
	543	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
	544	$network_to_use
	545	$data_limit
	546	$notification
	547
	548	%%END%%
	549
	550	if (( nodes > 0 ))
	551	then
	552
	553	cat >> $job_to_send << %%END%%
	554	# @ node = $nodes
	555	# @ tasks_per_node = $tasks_per_node
	556	# @ node_usage = $node_usage
	557	# @ queue
	558
	559	%%END%%
	560
	561	else
	562
	563	if [[ $remote_host != ibmy ]]
	564	then
	565
	566	cat >> $job_to_send << %%END%%
	567	# @ blocking = unlimited
	568	# @ total_tasks = $numprocs
	569	# @ node_usage = $node_usage
	570	# @ queue
	571
	572	%%END%%
	573
	574	else
	575
	576	cat >> $job_to_send << %%END%%
	577	# @ node = 1
	578	# @ total_tasks = $numprocs
	579	# @ queue
	580
	581	%%END%%
	582
	583	fi
	584
	585	fi
	586
	587	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	588	then
	589
	590	cat > $job_to_send << %%END%%
	591	#!/bin/ksh
	592
	593	# @ job_type = serial
	594	# @ node_usage = $node_usage
	595	# @ job_name = palm
	596	# @ wall_clock_limit = ${timestring},$timestring
	597	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	598	# @ output = $remote_dayfile
	599	# @ error = $remote_dayfile
	600	# @ class = $queue
	601	$notification
	602
	603	# @ queue
	604
	605	%%END%%
	606
	607	elif [[ $remote_host = nech ]]
	608	then
	609
	610	if (( nodes > 1 ))
	611	then
	612	cat > $job_to_send << %%END%%
	613	#!/bin/ksh
	614	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	615	#PBS -l ${qsubmem}=${Memory}gb
	616	#PBS -b $nodes
	617	#PBS -o $remote_dayfile
	618	#PBS -N palm
	619	#PBS -j o
	620	#PBS -T mpisx
	621
	622	%%END%%
	623
	624	elif [[ $numprocs != 0 ]]
	625	then
	626	cat > $job_to_send << %%END%%
	627	#!/bin/ksh
	628	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	629	#PBS -l ${qsubmem}=${Memory}gb
	630	#PBS -o $remote_dayfile
	631	#PBS -N palm
	632	#PBS -j o
	633
	634	%%END%%
	635
	636	else
	637	cat > $job_to_send << %%END%%
	638	#!/bin/ksh
	639	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	640	#PBS -o $remote_dayfile
	641	#PBS -j o
	642
	643	%%END%%
	644
	645	fi
	646
[35]	647	elif [[ $remote_host = neck ]]
	648	then
	649
	650	if (( nodes > 1 ))
	651	then
	652	cat > $job_to_send << %%END%%
	653	#!/bin/ksh
	654	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	655	#PBS -l ${qsubmem}=${Memory}gb
	656	#PBS -b $nodes
	657	#PBS -o $remote_dayfile
	658	#PBS -N palm
	659	#PBS -j o
	660	#PBS -T mpisx
	661
	662	%%END%%
	663
	664	elif [[ $numprocs != 0 ]]
	665	then
	666	cat > $job_to_send << %%END%%
	667	#!/bin/ksh
	668	#PBS -q ${queue}
	669	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
	670	#PBS -l ${qsubmem}=${Memory}gb
	671	#PBS -o $remote_dayfile
	672	#PBS -N palm
	673	#PBS -j o
	674	#PBS -T mpisx
	675
	676	%%END%%
	677
	678	else
	679	cat > $job_to_send << %%END%%
	680	#!/bin/ksh
	681	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	682	#PBS -o $remote_dayfile
	683	#PBS -j o
	684
	685	%%END%%
	686
	687	fi
	688
[1]	689	elif [[ $remote_host = lctit ]]
	690	then
	691	cat > $job_to_send << %%END%%
	692	#!/bin/bash
	693	#$ -S /bin/bash
	694	cd $job_catalog
[69]	695	export PATH=$PALM_BIN:\$PATH
	696	echo \$PATH
[1]	697
	698	%%END%%
	699
	700	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
	701	submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
	702
	703	else
	704
	705	cat > $job_to_send << %%END%%
	706	# @\$-q ${queue}
	707	# @\$-l${qsubtime} $timestring
	708	# @\$-l${qsubmem} ${memory}mb
	709	# @\$-o $remote_dayfile
	710	# @\$-eo
	711
	712	%%END%%
	713
	714	fi
	715
	716
	717	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	718	# BEI EXIT VERANLASSEN
	719	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	720	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	721	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	722	then
	723	echo "set +vx" >> $job_to_send
	724	echo "trap '" >> $job_to_send
	725	echo "set +vx" >> $job_to_send
[35]	726	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
[1]	727	then
	728	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
	729	then
	730	return_queue=c1
	731	elif [[ $remote_host = ibms ]]
	732	then
	733	return_queue=p_normal
	734	elif [[ $remote_host = ibmy ]]
	735	then
	736	return_queue=serial
	737	elif [[ $remote_host = lctit ]]
	738	then
	739	return_queue=default
[35]	740	elif [[ $remote_host = neck ]]
[1]	741	then
[35]	742	return_queue=S
[1]	743	else
	744	return_queue=unknown
	745	fi
	746
	747	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	748	then
	749
	750	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	751	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	752	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	753	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	754	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	755	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	756	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	757	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	758	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	759	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	760
	761	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	762	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	763
	764	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	765	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
	766	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	767
	768	elif [[ $remote_host = nech ]]
	769	then
	770	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	771	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	772	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	773	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	774	echo "#PBS -j o" >> $job_to_send
	775	echo " " >> $job_to_send
	776	echo "set -x" >> $job_to_send
	777	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	778	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	779	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	780	echo "%%END%%" >> $job_to_send
	781
[35]	782	elif [[ $remote_host = neck ]]
	783	then
	784	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	785	echo "#PBS -q $return_queue" >> $job_to_send
	786	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	787	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	788	echo "#PBS -j o" >> $job_to_send
	789	echo " " >> $job_to_send
	790	echo "set -x" >> $job_to_send
	791	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	792	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	793	echo "%%END%%" >> $job_to_send
	794
[1]	795	elif [[ $remote_host = lctit ]]
	796	then
	797	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	798	echo "set -x" >> $job_to_send
	799	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
	800	echo "cd $job_catalog" >> $job_to_send
	801	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	802	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	803	echo "%%END%%" >> $job_to_send
	804
	805	else
	806
	807	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	808	echo "# @\\\$-q $return_queue" >> $job_to_send
	809	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	810	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	811	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	812	then
	813	echo "# @\$-l mpp_p=0" >> $job_to_send
	814	fi
	815	echo '# @\$-lF 10mb' >> $job_to_send
	816	# echo '# @\$-o /dev/null' >> $job_to_send
	817	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	818	echo '# @\\\$-eo' >> $job_to_send
	819	echo " " >> $job_to_send
	820	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	821	then
	822	echo "set +vx" >> $job_to_send
	823	echo ". .profile" >> $job_to_send
	824	fi
	825	echo "set -x" >> $job_to_send
	826	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	827	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	828	echo "%%END%%" >> $job_to_send
	829	fi
	830
	831	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	832	then
	833	echo "llsubmit scpjob.$kennung" >> $job_to_send
	834	elif [[ $remote_host = lctit ]]
	835	then
	836	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	837	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
	838	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
	839	echo "sleep 10" >> $job_to_send
	840	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	841	then
	842	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	843	elif [[ $remote_host = t3es ]]
	844	then
	845	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
	846	else
	847	echo "qsub scpjob.$kennung" >> $job_to_send
	848	fi
	849	echo "rm scpjob.$kennung" >> $job_to_send
	850	if [[ $remote_host = nech ]]
	851	then
	852	echo "cd -" >> $job_to_send
	853	fi
	854	else
	855	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	856	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	857	fi
	858	echo "set -x" >> $job_to_send
	859	echo " ' exit" >> $job_to_send
	860	echo "set -x" >> $job_to_send
	861	fi
	862
	863
	864
	865	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	866	cat $file_to_send >> $job_to_send
[69]	867
[1]	868	if [[ $remote_host = ibm ]]
	869	then
	870	echo " " >> $job_to_send
	871	echo "exit" >> $job_to_send
	872	fi
[69]	873	if [[ $remote_host = lctit ]]
	874	then
	875	echo " " >> $job_to_send
	876	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	877	fi
[1]	878
	879
	880
	881	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	882	if [[ -z $remote_user ]]
	883	then
	884	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	885	then
	886	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	887	fi
	888	fi
	889
	890
	891
	892	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	893	if [[ $no_submit = false ]]
	894	then
	895	if [[ $remote_host != $local_host ]]
	896	then
	897	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	898	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	899	then
[82]	900	job_catalog_save=$job_catalog
	901	job_catalog=job_queue
	902	elif [[ $remote_host = nech ]]
	903	then
	904	job_catalog_save=$job_catalog
	905	job_catalog=/hpf/b/${remote_user}/job_queue
	906	fi
	907	if [[ $local_host = decalpha ]]
	908	then
	909	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	910	# PASSWORT AB
	911	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	912	elif [[ $remote_host = nech ]]
	913	then
	914	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	915	# TRANSFERIERT WERDEN
	916	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	917	else
[82]	918	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	919	fi
[82]	920	if [[ $? = 1 ]]
	921	then
	922	locat=scp; exit
	923	fi
	924	if [[ $remote_host = ibms ]]
	925	then
	926	job_catalog=$job_catalog_save
	927	fi
[1]	928	[[ $verify = true ]] && printf "\n >>> finished\n"
	929	else
	930	eval job_catalog=$job_catalog
	931	cp $job_to_send ${job_catalog}/$job_on_remhost
	932	fi
	933
	934
	935
	936	# NQS- BZW. LOADLEVELER-JOB STARTEN
	937	if [[ $remote_host != $local_host ]]
	938	then
	939	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[82]	940	if [[ $remote_host != lctit ]]
[1]	941	then
[82]	942	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[1]	943	else
[82]	944	# TIT ERLAUBT NUR DIE AUSFÜHRUNG GANZ BESTIMMTER KOMMANDOS
	945	# MIT SSH, DESHALB AUFRUF PER PIPE
	946	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
	947	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	948	echo "first try" > send_protocol
	949	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
	950	do
	951	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
	952	then
	953	printf "\n trying to submit job to TIT cluster..."
	954	else
	955	printf "\n +++ failed ... trying to submit again ..."
	956	fi
	957	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
	958	cat send_protocol
	959	done
	960	sleep 10
	961	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
	962	echo "$submcom $job_on_remhost"
	963	rm send_protocol
	964	fi
[1]	965
	966	[[ $verify = true ]] && printf " >>> o.k.\n"
	967	else
	968	cd $job_catalog
	969	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
	970	then
	971	eval $submcom $job_on_remhost
	972	elif [[ $local_host = lctit ]]
	973	then
	974	eval $submcom $job_on_remhost
	975	echo "$submcom $job_on_remhost"
	976	chmod u+x $job_on_remhost
	977	elif [[ $local_host = nech ]]
	978	then
	979	if [[ $queue = default ]]
	980	then
	981	eval $submcom $job_on_remhost
	982	else
	983	eval $submcom -q $queue $job_on_remhost
	984	fi
	985	else
	986	qsub $job_on_remhost
	987	fi
[69]	988	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
	989	[[ $local_host != lctit ]] && rm $job_on_remhost
[1]	990	cd - > /dev/null
	991	fi
	992	fi
	993
	994
	995
	996	# ABSCHLUSSARBEITEN
	997
	998	if [[ $no_submit = false ]]
	999	then
	1000	rm $job_to_send
	1001	fi
	1002	[[ $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 |