Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 736

Last change on this file since 736 was 717, checked in by gryschka, 14 years ago
added lckordi
Property svn:keywords set to `Id Rev`
File size: 49.3 KB

Rev	Line
[1]	1	#!/bin/ksh
[82]	2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
[169]	3	# $Id: subjob 717 2011-04-03 16:04:01Z suehring $
[1]	4
	5	# Prozedur zum automatischen Generieren von Batch-Jobs, die unter NQS
	6	# laufen sollen und deren Ergebnis (Dayfile) zum Job-generierenden
	7	# Host zurueckgeschickt werden sollen
	8
	9
	10	# letzte Aenderung:
	11	# 29/06/94 - Siggi - Beginn mit Erstellung der Originalversion
	12	# 08/07/94 - Siggi - Originalversion abgeschlossen (Version 1.0)
	13	# 06/02/98 - Siggi - berte validiert
	14	# 27/01/01 - Siggi - ground.yonsei.ac.kr validiert, Jobs zur T3E in Korea
	15	# moeglich
	16	# 08/02/01 - Siggi - alle subjob-Meldungen ins englische uebersetzt
	17	# 25/05/02 - Siggi - Unterstuetzung des LoadLeveler
	18	# 30/05/02 - Siggi - Validierung fuer ibm-Rechner in Seoul (nobel) sowie
	19	# allgemeine Anpassungen fuer ibm-Rechner
	20	# 15/10/02 - Siggi - Neue Default-Jobklasse (p_normal) fuer IBM in Seoul
	21	# Ruecktransfer des Jobprotokolls fuer diese
	22	# Maschine abgeschaltet
	23	# 31/10/02 - Siggi - berni validiert
	24	# 06/11/02 - Siggi - Neue Jobklassen auf ibmb und ibmh
	25	# 08/11/02 - Siggi - quanero validiert
	26	# 11/12/02 - Siggi - Notification fuer Transfer-Jobs abgeschaltet
	27	# 23/01/03 - Siggi - hostname nobel changed to nobela
	28	# 06/02/03 - Siggi - gregale validated
	29	# 12/02/03 - Siggi - orkan and maestro validated
	30	# 21/02/03 - Siggi - all nobel nodes in Seoul validated
	31	# 12/03/03 - Siggi - nec at DKRZ validated
	32	# 13/03/03 - Siggi - new nqs resource variable Memory
	33	# 07/04/03 - Siggi - processor request option -c on nech needs tasks per
	34	# node
	35	# 11/04/03 - Siggi - network on ibms has attribute "shared"
	36	# 31/07/03 - Siggi - nqs2 on nech implemented (provisional: -h nech2)
	37	# cxxl added to ibmh
	38	# 29/08/03 - Siggi - changes in job queues and communication system on
	39	# ibms
	40	# 24/10/03 - Siggi - using alternate hanni address 130.75.4.2
	41	# 30/10/03 - Siggi - nech is not supported any more
	42	# 10/11/03 - Siggi - nech2 renamed to nech
	43	# 20/11/03 - Siggi - submit command on nech changed from qsub.test to qsub
	44	# 29/03/04 - Siggi - ground not supported any more, gfdl3 validated
	45	# 31/03/04 - Siggi - new option -N for node usage
	46	# 12/04/04 - Siggi - scp2 instead of scp used for transfer from decalpha
	47	# due to error in ssh installation (otherwise a prompt
	48	# for the password appears)
	49	# 23/07/04 - Siggi - changes due to the new berni configuration
	50	# (federation switch)
	51	# 01/09/04 - Gerald new job-classes on hanni
	52	# 08/09/04 - Siggi - hanni IP address changed to 130.75.4.10
	53	# 23/11/04 - Siggi - new job class cdata on hanni and berni
	54	# 03/12/04 - Siggi - notification on ibm switched of in case of
	55	# delete_dayfile = true, node usage in cdev set to
	56	# shared
	57	# 16/02/05 - Gerald hababai validated
	58	# 29/03/05 - Micha - new job class channi on hanni
	59	# 11/05/05 - Siggi - ConsumableMemory is now required as resource keyword
	60	# on ibms
	61	# 24/05/05 - Siggi - Default queue on ibms changed from p_normal_1.3 to
	62	# p_normal
	63	# 30/06/05 - Siggi - network changed for queue cdev from "us" to "ip"
	64	# 12/07/05 - Siggi - in network.mpi on ibmh/ibmb "csss" changed to
	65	# "sn_all", new job class cexp
	66	# 08/09/05 - Siggi - IP-address of gfdl3 changed
	67	# 31/10/05 - Siggi - new job class pp on hurricane, serial jobs on
	68	# hurricane (with -X0)
	69	# 01/11/05 - Siggi - missing queue for jobs submitted on nech (for nech)
	70	# added
	71	# 30/12/05 - Siggi - change of IP adresses in subnet 130.75.105
	72	# 09/02/06 - Siggi - ibmy admitted
	73	# 10/02/06 - Siggi - scp2 changed to /bin/scp on decalpha
	74	# 13/04/06 - Siggi - ostria admitted
	75	# 18/04/06 - Siggi - new option -O for OpenMP usage
	76	# 24/05/06 - Siggi - lctit admitted, ftpjob renamed scpjob
	77	# 25/07/06 - Siggi - gfdl5 (ibmy) admitted for submitting jobs
	78	# 27/09/06 - Siggi - breg/hreg extended with berni/hanni
	79	# 25/10/06 - Siggi - data_limit set to 1.76 GByte on hanni and berni
	80	# 28/11/06 - Siggi - levanto admitted
[5]	81	# 13/02/07 - Siggi - hpmuk releated code removed
[35]	82	# 01/03/07 - Siggi - adjustments for RIAM machines gate and NEC-SX8 (n-sx)
[82]	83	# 12/04/07 - Siggi - option -f (filetransfer protocol) removed, scp only
[102]	84	# 27/07/07 - Siggi - autan admitted
[108]	85	# 03/08/07 - Marcus- lcfimm admitted
[113]	86	# 08/10/07 - Siggi - further job classes added for hanni (csoloh...)
[118]	87	# 15/10/07 - Siggi - preliminary adjustments for lctit based on Jin's
	88	# suggestions
[122]	89	# 19/10/07 - Marcus- add new optional argument -g group_number
[125]	90	# 19/10/07 - Siggi - a ";" was missing in the last change done by Marcus
[129]	91	# 30/10/07 - Marcus- further adjustments for queues on lctit
[164]	92	# 15/05/08 - Siggi - adjustments for lcxt4 (Bergen Center for Computational
	93	# Science)
[179]	94	# 14/07/08 - Siggi - adjustments for lcsgih
[205]	95	# 23/09/08 - Gerald- paesano admitted
[206]	96	# 02/10/08 - Siggi - PBS adjustments for lcxt4
[251]	97	# 02/03/09 - Siggi - Adjustments for new NEC-SX9 at RIAM
[292]	98	# 16/04/09 - Marcus- Adjustments for lcsgib and lcsgih
[312]	99	# 21/04/09 - Siggi - adjustments for new IBM at DKRZ, which is now ibmh
[325]	100	# 18/05/09 - Siggi - Settings for serial jobs on lcsgi changed
[343]	101	# 24/06/09 - BjornM- adjustments for lcxt4 (loading modules manually)
[352]	102	# 08/07/09 - Siggi - option -e added (email notification on lcsgih/b)
[366]	103	# 20/07/09 - Siggi - On lcsgi, jobs for returning the job protocol are
	104	# now run on the data nodes (feature=data)
[367]	105	# 25/08/09 - BjornM- adapted for lck
[372]	106	# 26/08/09 - Marcus- caurus admitted; optional qos feature hiprio on lcsgi
[376]	107	# 03/09/09 - Siggi - PBS sgi feature directive only used if explicitly
	108	# set in the config file by the user
[416]	109	# 16/10/09 - Carolin-adjustments for archiving on SGI-ICE of binary files;
	110	# adding special1q
[407]	111	# 01/12/09 - BjornM- re-adjustments for lcxt4, added sno (130.75.105.113)
[416]	112	# 16/10/09 - Carolin-adding permq
[440]	113	# 01/02/10 - Siggi - adapted for lcxt5m and lckyoto (Fujitsu HX600)
[475]	114	# 03/02/10 - Siggi - bug in serial jobs removed
[492]	115	# 26/02/10 - BjornM- re-adjustments for lcxt4 (new modules, email
	116	# notification)
[493]	117	# 01/03/10 - Siggi - loading of modules controlled by environment variable
	118	# module_calls
[551]	119	# 17/08/10 - BjornM- re-adjustments for lcxt4 (location of qsub)
	120	# 25/08/10 - BjornM- account geofysisk replaced by guest for lcxt4
[552]	121	# 25/08/10 - Siggi - new variable project_account in pbs-statements for
	122	# lcxt4
[622]	123	# 08/12/10 - Siggi - initialization of the module command changed for
	124	# SGI-ICE/lcsgi
	125	# adjustments for Kyushu Univ. (lcrte, ibmku)
[635]	126	# 14/12/10 - Siggi - adjustments for new Tsubame system at Tokyo
	127	# institute of technology (lctit)
[678]	128	# 02/02/11 - Siggi - further asjustments for Tsubame concerning openMP
[693]	129	# 06/03/11 - Siggi - adjustments for ibmkisti
[696]	130	# 17/03/11 - Siggi - adjustments for openmp usage on ibmkisti
[717]	131	# 03/04/11 - Micha - added lckordi
[1]	132
[437]	133
[552]	134
[1]	135	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
[352]	136	delete_dayfile=false
	137	email_notification=none
[122]	138	group_number=none
[1]	139	locat=normal
	140	no_default_queue=none
	141	no_submit=false
	142	job_catalog="~/job_queue"
	143	job_name=none
	144	local_user=$LOGNAME
	145	node_usage=shared
[475]	146	numprocs=0
[1]	147	punkte="..........................................................."
	148	submcom=qsub
	149	queue=default
	150	remote_host=none
	151	remote_user=""
	152	verify=true
	153
	154	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	155	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	156	typeset -L20 spalte1
	157	typeset -R40 spalte2
	158	typeset -L60 spalte3
	159
	160
	161
	162	# FEHLERBEHANDLUNG
	163	# BEI EXIT:
	164	trap 'if [[ $locat != normal ]]
	165	then
	166	case $locat in
	167	(option) printf "\n --> available optios can be displayed"
	168	printf " by typing:"
	169	printf "\n \"subjob ?\" \n";;
	170	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	171	(*) printf "\n +++ unknown error"
	172	printf "\n please inform S. Raasch!\n"
	173	esac
	174	[[ -f $job_to_send ]] && rm $job_to_send
	175	printf "\n\n+++ SUBJOB killed \n\n"
	176	fi' exit
	177
	178
	179	# BEI TERMINAL-BREAK:
	180	trap '[[ -f $job_to_send ]] && rm $job_to_send
	181	printf "\n\n+++ SUBJOB killed \n\n"
	182	exit
	183	' 2
	184
	185
	186
	187
	188	# LOKALEN HOSTNAMEN FESTSTELLEN
	189	local_host=$(hostname)
	190
	191
	192
	193	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
	194	# UEBERHAUPT ZULAESSIG IST
[717]	195	# Note: One of the entries for "lck" or "lckordi" always has to be
	196	# comment out, because the hostname (node*) is same for both machines
[1]	197	case $local_host in
	198	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[102]	199	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	200	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[400]	201	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
	202	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
	203	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
	204	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
	205	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
[291]	206	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[12]	207	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[292]	208	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[372]	209	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[693]	210	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
[108]	211	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
[1]	212	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	213	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[108]	214	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
[367]	215	(node*) local_addres=165.132.26.61 local_host=lck;;
[717]	216	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
[693]	217	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
[101]	218	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[1]	219	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	220	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	221	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	222	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[164]	223	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
[400]	224	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
	225	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
	226	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
	227	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
[292]	228	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
[210]	229	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
[440]	230	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
[312]	231	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[5]	232	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
[407]	233	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
[5]	234	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
	235	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[437]	236	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
[164]	237	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
[1]	238	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	239	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	240	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	241	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[5]	242	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[622]	243	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
[1]	244	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
	245	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
[259]	246	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
[181]	247	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
[196]	248	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
[635]	249	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
[1]	250	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
[622]	251	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
[1]	252	(*) printf "\n +++ \"$local_host\" unknown";
	253	printf "\n please inform S. Raasch!";
	254	locat=parameter; exit;;
	255	esac
	256
	257
	258
	259	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	260	remote_host=$local_host
	261
	262
	263
	264
	265	# PROZEDUROPTIONEN EINLESEN
[352]	266	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
[1]	267	do
	268	case $option in
	269	(c) job_catalog=$OPTARG;;
	270	(d) delete_dayfile=true;;
	271	(D) no_submit=true;;
[352]	272	(e) email_notification=$OPTARG;;
[125]	273	(g) group_number=$OPTARG;;
[1]	274	(h) remote_host=$OPTARG;;
	275	(m) memory=$OPTARG;;
	276	(n) job_name=$OPTARG;;
	277	(N) node_usage=$OPTARG;;
	278	(O) threads_per_task=$OPTARG;;
	279	(q) no_default_queue=$OPTARG;;
	280	(t) cputime=$OPTARG;;
	281	(T) tasks_per_node=$OPTARG;;
	282	(u) remote_user=$OPTARG;;
	283	(v) verify=false;;
	284	(X) numprocs=$OPTARG;;
	285	(\?) printf "\n +++ Option $OPTARG unknown \n";
	286	locat=option; exit;;
	287	esac
	288	done
	289
	290
	291	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	292	shift OPTIND-1; file_to_send=$1
	293
	294
	295	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	296	if [ "$1" = "?" ]
	297	then
	298	(printf "\n *** subjob can be called as follows:\n"
	299	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	300	printf "\n Description of available options:\n"
	301	printf "\n Option Description Default-Value"
	302	printf "\n -c job-input- and output-catalog ~/job_queue"
	303	printf "\n -d no job-protocol will be created ---"
	304	printf "\n -D only the job-file will be created ---"
	305	printf "\n -h execution host, available hosts: $remote_host"
[693]	306	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
[251]	307	printf "\n lctit, nech, necriam, unics"
[1]	308	printf "\n -m memory demand per process in MByte ---"
	309	printf "\n -n jobname <jobdatei>"
	310	printf "\n -O threads per task (for OpenMP usage) 1"
	311	printf "\n -q job-queue to be used default"
	312	printf "\n -t allowed cpu-time in seconds ---"
	313	printf "\n -T tasks per node (on parallel hosts) ---"
	314	printf "\n -u username on execution host from .netrc"
	315	printf "\n -v no prompt for confirmation ---"
	316	printf "\n -X # of processors (on parallel hosts) 1"
	317	printf "\n "
	318	printf "\n The only possible positional parameter is <jobfile>:"
	319	printf "\n The complete NQS-job must be provided here."
	320	printf "\n <jobfile>=? creates this outline\n\n") \| more
	321	exit
	322	fi
	323
	324
	325
	326	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	327	if [[ "$file_to_send" = "" ]]
	328	then
	329	printf "\n +++ job-file missing"
	330	locat=parameter; exit
	331	else
	332	if [[ -f $file_to_send ]]
	333	then
	334	true
	335	else
	336	printf "\n +++ job-file: "
	337	printf "\n $file_to_send"
	338	printf "\n does not exist"
	339	locat=parameter; exit
	340	fi
	341	fi
	342
	343
	344
	345	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	346	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	347	if [[ $job_name = none ]]
	348	then
	349	job_name=$file_to_send
	350	fi
	351	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	352	then
	353	printf "\n +++ job-file name: "
	354	printf "\n $job_name"
	355	printf "\n must not contain \"/\"-characters"
	356	locat=parameter; exit
	357	fi
	358
	359
	360
	361
	362	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	363	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	364	if [[ $remote_host = none ]]
	365	then
	366	printf "\n +++ host missing"
	367	locat=option; exit
	368	else
	369	case $remote_host in
[251]	370	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[312]	371	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[693]	372	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[622]	373	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
[251]	374	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	375	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	376	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
[440]	377	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
[540]	378	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
	379	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
[367]	380	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[717]	381	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
[635]	382	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
[544]	383	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.4.9-snap.201005191035/bin/qsub;;
[437]	384	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
[693]	385	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
[251]	386	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	387	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
	388	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	389	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	390	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	391	locat=parameter; exit;;
[1]	392	esac
	393	fi
	394
	395
	396	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	397	if [[ $no_default_queue != none ]]
	398	then
	399	error=false
	400	ndq=$no_default_queue
	401	case $remote_host in
	402	(ibm) case $ndq in
	403	(p690_express\|p690_standard\|p690_long) error=false;;
	404	(*) error=true;;
	405	esac;;
	406	(ibmh) case $ndq in
[312]	407	(no_class) error=false;;
[1]	408	(*) error=true;;
	409	esac;;
[693]	410	(ibmkisti) case $ndq in
	411	(class.32plus\|class.1-2\|class.2-32) error=false;;
	412	(*) error=true;;
	413	esac;;
[622]	414	(ibmku) case $ndq in
	415	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	416	(*) error=true;;
	417	esac;;
[1]	418	(ibms) case $ndq in
	419	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	420	(*) error=true;;
	421	esac;;
	422	(ibmy) case $ndq in
	423	(parallel) error=false;;
	424	(*) error=true;;
	425	esac;;
[440]	426	(lckyoto) case $ndq in
	427	(eh\|ph) error=false;;
	428	(*) error=true;;
	429	esac;;
[191]	430	(lcsgib\|lcsgih) case $ndq in
[416]	431	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
[191]	432	(*) error=true;;
	433	esac;;
[1]	434	(lctit) case $ndq in
[635]	435	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	436	(*) error=true;;
	437	esac;;
	438	(t3eb) case $ndq in
	439	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	440	(*) error=true;;
	441	esac;;
[251]	442	(necriam) case $ndq in
[253]	443	(SP\|SS\|P6) error=false;;
[251]	444	(*) error=true;;
	445	esac;;
[1]	446	(t3eh) case $ndq in
	447	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	448	(*) error=true;;
	449	esac;;
	450	(t3ej2\|t3ej5) case $ndq in
	451	(low\|normal\|high) error=false;;
	452	(*) error=true;;
	453	esac;;
	454	(t3es) case $ndq in
	455	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	456	(*) error=true;;
	457	esac;;
	458	(unics) case $ndq in
	459	(unics\|ht) error=false;;
	460	(*) error=true;;
	461	esac;;
	462	esac
	463	if [[ $error = true ]]
	464	then
	465	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	466	locat=parameter; exit
	467	else
	468	queue=$no_default_queue
	469	fi
	470	fi
	471
	472
	473
	474	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	475	# AUFTEILEN
	476	done=false
	477	while [[ $done = false ]]
	478	do
	479	if (( $cputime <= 0 ))
	480	then
	481	printf "\n +++ wrong cpu-time or cpu-time missing"
	482	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	483	printf "\n >>> "
	484	read cputime 1>/dev/null 2>&1
	485	else
	486	done=true
	487	fi
	488	done
	489	if [[ $remote_host = nech ]]
	490	then
	491	if (( tasks_per_node != 0 ))
	492	then
	493	(( cputime = cputime * tasks_per_node ))
	494	elif [[ $numprocs != 0 ]]
	495	then
	496	(( cputime = cputime * numprocs ))
	497	fi
	498	fi
	499	(( stunden = cputime / 3600 ))
	500	(( resttime = cputime - stunden * 3600 ))
	501	(( minuten = resttime / 60 ))
	502	(( sekunden = resttime - minuten * 60 ))
	503	timestring=${stunden}:${minuten}:${sekunden}
	504
	505
	506
	507	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	508	done=false
	509	while [[ $done = false ]]
	510	do
	511	if (( memory <= 0 ))
	512	then
	513	printf "\n +++ wrong memory demand or memory demand missing"
	514	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	515	printf "\n >>> "
	516	read memory 1>/dev/null 2>&1
	517	else
	518	done=true
	519	fi
	520	done
	521
[251]	522	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
[1]	523	then
	524	if (( tasks_per_node != 0 ))
	525	then
	526	(( Memory = memory * tasks_per_node / 1000 ))
	527	elif [[ $numprocs != 0 ]]
	528	then
	529	(( Memory = memory * numprocs / 1000 ))
	530	else
	531	(( Memory = memory / 1000 ))
	532	fi
[635]	533	elif [[ $remote_host = lctit ]]
	534	then
	535	(( Memory = memory * tasks_per_node / 1000 ))
[1]	536	fi
	537
	538
	539	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	540	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	541	then
	542	(( memory = memory * threads_per_task ))
	543	fi
	544
	545
	546	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	547	if (( tasks_per_node != 0 ))
	548	then
[622]	549	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
[1]	550	fi
	551
[622]	552	# Calculate number of processes per node
	553	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	554
[696]	555	# Calculate number of MPI tasks
	556	(( mpi_tasks = numprocs / threads_per_task ))
[1]	557
[696]	558
[1]	559	# HEADER-AUSGABE
	560	if [[ $verify = true ]]
	561	then
	562	printf "\n\n"
	563	printf "#--------------------------------------------------------------# \n"
	564	spalte1=SUBJOB;spalte2=$(date)
	565	printf "\| $spalte1$spalte2 \| \n"
	566	printf "\| \| \n"
	567	printf "\| values of parameters/options: \| \n"
	568	spalte1=local_host$punkte; spalte2=$punkte$local_host
	569	printf "\| $spalte1$spalte2 \| \n"
	570	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	571	printf "\| $spalte1$spalte2 \| \n"
	572	spalte1=queue$punkte; spalte2=$punkte$queue
	573	printf "\| $spalte1$spalte2 \| \n"
	574	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	575	printf "\| $spalte1$spalte2 \| \n"
	576	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	577	printf "\| $spalte1$spalte2 \| \n"
	578	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	579	printf "\| $spalte1$spalte2 \| \n"
	580	printf "#--------------------------------------------------------------# \n\n"
	581
	582
	583	# KONTROLLABFRAGE, OB ALLES O.K.
	584	antwort="dummy"
	585	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	586	do
	587	read antwort?" >>> continue (y/n) ? "
	588	done
	589	if [[ $antwort = n \|\| $antwort = N ]]
	590	then
	591	locat=verify; exit
	592	fi
	593	printf "\n"
	594	fi
	595
	596
	597
	598	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	599	kennung=$RANDOM
	600	job_on_remhost=${job_name}_${kennung}_$local_host
	601	job_to_send=job_to_send_$kennung
	602	if [[ $delete_dayfile = false ]]
	603	then
	604	remote_dayfile=${local_host}_${job_name}_result_$kennung
	605	local_dayfile=${remote_host}_${job_name}
	606	else
	607	remote_dayfile=/dev/null
	608	fi
	609
	610
	611
[622]	612	# Generate the batch job scripts (qsub/msub/LoadLeveler)
[1]	613	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	614	then
	615
[622]	616	# General LoadLeveler settings
	617	execute_in_shell="#!/bin/ksh"
	618	use_shell="# @ shell = /bin/ksh"
	619	consumable_memory="ConsumableMemory($memory mb)"
	620	class="# @ class = $queue"
	621	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	622	network_to_use="# @ network.mpi = sn_all,shared,us"
	623	data_limit="# @ data_limit = 1.76gb"
	624	image_size="# @ image_size = 50"
[693]	625	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	626
[693]	627	if [[ $email_notification = none ]]
	628	then
	629	notify_user=""
	630	else
	631	notify_user="# @ notify_user = $email_notification"
	632	if [[ $delete_dayfile = true ]]
	633	then
	634	notification='# @ notification = never'
	635	fi
	636	fi
[622]	637
[312]	638	if [[ $remote_host = ibmh ]]
[1]	639	then
[312]	640	data_limit=""
	641	network_to_use=""
	642	class=""
	643	environment=""
[693]	644	elif [[ $remote_host = ibmkisti ]]
	645	then
	646	network_to_use="# @ network.MPI = sn_all,shared,US"
	647	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	648	if [[ $threads_per_task = 1 ]]
	649	then
	650	rset="# @ rset = RSET_MCM_AFFINITY"
	651	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	652	fi
[693]	653	environment=""
	654	use_shell=""
	655	data_limit=""
	656	image_size=""
[622]	657	elif [[ $remote_host = ibmku ]]
	658	then
	659	execute_in_shell="#!/usr/bin/ksh"
	660	use_shell="# @ shell = /usr/bin/ksh"
	661	consumable_memory=""
	662	environment=""
	663	network_to_use="# @ network.mpi = sn_all,shared,us"
	664	data_limit=""
	665	image_size=""
	666	elif [[ $remote_host = ibms ]]
	667	then
	668	network_to_use="# @ network.mpi = csss,shared,us"
	669	elif [[ $remote_host = ibmy ]]
	670	then
	671	consumable_memory=""
	672	network_to_use=""
[1]	673	fi
	674
	675	cat > $job_to_send << %%END%%
[622]	676	$execute_in_shell
	677	$use_shell
[1]	678
	679	# @ job_type = parallel
[693]	680	# @ job_name = $job_name
[1]	681	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	682	# @ output = $remote_dayfile
	683	# @ error = $remote_dayfile
[693]	684	$wall_clock_limit
[622]	685	$image_size
[312]	686	$class
	687	$environment
[1]	688	$network_to_use
	689	$data_limit
[693]	690	$rset
	691	$mcm_affinity_options
[1]	692	$notification
[693]	693	$notify_user
[1]	694
	695	%%END%%
	696
	697	if (( nodes > 0 ))
	698	then
	699
[693]	700	if [[ $remote_host != ibmkisti ]]
	701	then
	702
	703	cat >> $job_to_send << %%END%%
[1]	704	# @ node = $nodes
[622]	705	# @ tasks_per_node = $processes_per_node
[1]	706	# @ node_usage = $node_usage
	707	# @ queue
	708
	709	%%END%%
	710
[693]	711	else
	712
	713	cat >> $job_to_send << %%END%%
[696]	714	# @ total_tasks = $mpi_tasks
[693]	715	# @ blocking = unlimited
	716	# @ queue
	717
	718	%%END%%
	719
	720	fi
	721
[1]	722	else
	723
	724	if [[ $remote_host != ibmy ]]
	725	then
	726
	727	cat >> $job_to_send << %%END%%
	728	# @ blocking = unlimited
	729	# @ total_tasks = $numprocs
	730	# @ node_usage = $node_usage
	731	# @ queue
	732
	733	%%END%%
	734
	735	else
	736
	737	cat >> $job_to_send << %%END%%
	738	# @ node = 1
	739	# @ total_tasks = $numprocs
	740	# @ queue
	741
	742	%%END%%
	743
	744	fi
	745
	746	fi
	747
[696]	748	# workaround because of silly job filter on ibmkisti
	749	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	750	then
	751	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	752	fi
	753
[1]	754	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	755	then
	756
	757	cat > $job_to_send << %%END%%
	758	#!/bin/ksh
	759
	760	# @ job_type = serial
	761	# @ node_usage = $node_usage
	762	# @ job_name = palm
	763	# @ wall_clock_limit = ${timestring},$timestring
	764	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	765	# @ output = $remote_dayfile
	766	# @ error = $remote_dayfile
[312]	767	$class
[1]	768	$notification
	769
	770	# @ queue
	771
	772	%%END%%
	773
[108]	774	elif [[ $remote_host = lcfimm ]]
	775	then
	776
	777	if [[ $numprocs != 0 ]]
	778	then
	779	cat > $job_to_send << %%END%%
	780	#!/bin/ksh
	781	#PBS -N $job_name
[552]	782	#PBS -A $project_account
[108]	783	#PBS -l walltime=$timestring
[622]	784	#PBS -l nodes=${nodes}:ppn=$processes_per_node
[108]	785	#PBS -l pmem=${memory}mb
	786	#PBS -m abe
	787	#PBS -o $remote_dayfile
	788	#PBS -j oe
	789	mpd &
	790
	791	%%END%%
	792
[168]	793	else
	794	cat > $job_to_send << %%END%%
	795	#!/bin/ksh
	796	#PBS -N $job_name
[552]	797	#PBS -A $project_account
[168]	798	#PBS -l walltime=$timestring
	799	#PBS -l ncpus=1
	800	#PBS -l pmem=${memory}mb
	801	#PBS -m abe
	802	#PBS -o $remote_dayfile
	803	#PBS -j oe
	804
	805	%%END%%
	806
	807	fi
	808
[717]	809	elif [[ $remote_host = lck \|\| $remote_host = lckordi ]]
[368]	810	then
	811
	812	if [[ $numprocs != 0 ]]
	813	then
	814	cat > $job_to_send << %%END%%
	815	#!/bin/ksh
	816	#PBS -N $job_name
	817	#PBS -l walltime=$timestring
	818	#PBS -l ncpus=$numprocs
	819	#PBS -l pmem=${memory}mb
	820	#PBS -o $remote_dayfile
	821	#PBS -j oe
	822
	823	mpd &
	824
	825	%%END%%
	826
	827	else
	828	cat > $job_to_send << %%END%%
	829	#!/bin/ksh
	830	#PBS -N $job_name
	831	#PBS -l walltime=$timestring
	832	#PBS -l ncpus=1
	833	#PBS -l pmem=${memory}mb
	834	#PBS -o $remote_dayfile
	835	#PBS -j oe
	836
	837	%%END%%
	838
	839	fi
	840
[693]	841	elif [[ $remote_host = lcyon ]]
	842	then
	843
	844	if [[ $numprocs != 0 ]]
	845	then
	846	cat > $job_to_send << %%END%%
	847	#!/bin/ksh
	848	#PBS -N $job_name
	849	#PBS -l walltime=$timestring
	850	#PBS -l ncpus=$numprocs
	851	#PBS -l pmem=${memory}mb
	852	#PBS -o $remote_dayfile
	853	#PBS -j oe
	854
	855	%%END%%
	856
	857	else
	858	cat > $job_to_send << %%END%%
	859	#!/bin/ksh
	860	#PBS -N $job_name
	861	#PBS -l walltime=$timestring
	862	#PBS -l ncpus=1
	863	#PBS -l pmem=${memory}mb
	864	#PBS -o $remote_dayfile
	865	#PBS -j oe
	866
	867	%%END%%
	868
	869	fi
	870
[181]	871	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
[179]	872	then
	873
[376]	874	if [[ "$sgi_feature" != "" ]]
	875	then
	876	feature_directive="#PBS -l feature=$sgi_feature"
	877	else
	878	feature_directive=""
	879	fi
[399]	880
[416]	881	if [[ $queue = dataq \|\| $queue = permq ]]
[399]	882	then
	883	feature_directive="#PBS -l feature=data"
	884	fi
[376]	885
[416]	886	if [[ $queue = testq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
[191]	887	then
	888	queue_directive="#PBS -q $queue"
	889	else
	890	queue_directive=""
	891	fi
	892
[352]	893	if [[ $email_notification = none ]]
	894	then
	895	email_directive=""
	896	else
	897	email_directive="#PBS -M $email_notification"
	898	fi
	899
[179]	900	if [[ $numprocs != 0 ]]
	901	then
	902	cat > $job_to_send << %%END%%
[181]	903	#!/bin/bash
[179]	904	#PBS -N $job_name
	905	#PBS -l walltime=$timestring
[622]	906	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[181]	907	#PBS -l naccesspolicy=$node_usage
[179]	908	#PBS -o $remote_dayfile
	909	#PBS -j oe
[376]	910	$feature_directive
[191]	911	$queue_directive
[352]	912	$email_directive
[179]	913
[622]	914	eval \`/sw/swdist/bin/modulesinit\`
	915	#. /usr/share/modules/init/bash
[493]	916	$module_calls
[374]	917
[181]	918	echo ld_library_path=\$LD_LIBRARY_PATH
[179]	919
	920	%%END%%
	921
	922	else
	923	cat > $job_to_send << %%END%%
	924	#PBS -S /bin/bash
	925	#PBS -N $job_name
	926	#PBS -l walltime=$timestring
	927	#PBS -l ncpus=1
	928	#PBS -o $remote_dayfile
[475]	929	#PBS -j oe
[400]	930	$queue_directive
	931	$email_directive
[179]	932
[622]	933	eval \`/sw/swdist/bin/modulesinit\`
	934	#. /usr/share/modules/init/bash
[493]	935	$module_calls
[475]	936
[179]	937	%%END%%
	938
	939	fi
	940
[164]	941	elif [[ $remote_host = lcxt4 ]]
	942	then
	943
	944	if [[ $numprocs != 0 ]]
	945	then
	946	cat > $job_to_send << %%END%%
	947	#!/bin/ksh
	948	#PBS -S /bin/ksh
	949	#PBS -N $job_name
[552]	950	#PBS -A $project_account
[206]	951	#PBS -j oe
[164]	952	#PBS -l walltime=$timestring
	953	#PBS -l mppwidth=${numprocs}
[622]	954	#PBS -l mppnppn=${processes_per_node}
[164]	955	#PBS -m abe
	956	#PBS -o $remote_dayfile
[492]	957	$email_directive
[164]	958
[343]	959	. /opt/modules/default/init/ksh
[493]	960	$module_calls
[343]	961
[164]	962	%%END%%
	963
[108]	964	else
	965	cat > $job_to_send << %%END%%
	966	#!/bin/ksh
[168]	967	#PBS -S /bin/ksh
[108]	968	#PBS -N $job_name
[552]	969	#PBS -A $project_account
[206]	970	#PBS -j oe
[108]	971	#PBS -l walltime=$timestring
	972	#PBS -l ncpus=1
	973	#PBS -l pmem=${memory}mb
	974	#PBS -m abe
[492]	975	$email_directive
[108]	976	#PBS -o $remote_dayfile
	977
[343]	978	. /opt/modules/default/init/ksh
[493]	979	$module_calls
[343]	980
[108]	981	%%END%%
	982
	983	fi
	984
[440]	985	elif [[ $remote_host = lckyoto ]]
[437]	986	then
	987
[440]	988	cat > $job_to_send << %%END%%
	989	#!/bin/ksh
	990	# @\$-o $remote_dayfile
	991	# @\$-eo -oi
	992	# @\$-lP 16
	993	# @\$-lp 1
	994	# @\$-lm 28gb -llm unlimited -ls unlimited
	995	# @\$-q $queue
	996	# @\$-Pvn abs_pack
	997	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	998	#. /thin/local/etc/setprofile/intel-11.0.sh
	999	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1000	. ~/.myprofile
	1001	#. /home2/t/t51254/palm/current_version/myprofile
	1002	#. /thin/apps/pgi/mpi.sh
	1003	#
	1004	env
	1005	#
	1006	set -x
	1007
	1008	%%END%%
	1009
	1010	elif [[ $remote_host = lcxt5m ]]
	1011	then
	1012
[437]	1013	if [[ $numprocs != 0 ]]
	1014	then
	1015	cat > $job_to_send << %%END%%
	1016	#!/bin/ksh
	1017	#PBS -S /bin/ksh
	1018	#PBS -N $job_name
	1019	#PBS -j oe
	1020	#PBS -l walltime=$timestring
	1021	#PBS -l mppwidth=${numprocs}
[622]	1022	#PBS -l mppnppn=${processes_per_node}
[437]	1023	#PBS -m abe
	1024	#PBS -o $remote_dayfile
	1025
	1026	. /opt/modules/default/init/ksh
[493]	1027	$module_calls
[437]	1028
	1029	%%END%%
	1030
	1031	else
	1032	cat > $job_to_send << %%END%%
	1033	#!/bin/ksh
	1034	#PBS -S /bin/ksh
	1035	#PBS -N $job_name
	1036	#PBS -j oe
	1037	#PBS -l walltime=$timestring
	1038	#PBS -l ncpus=1
	1039	#PBS -l pmem=${memory}mb
	1040	#PBS -m abe
	1041	#PBS -o $remote_dayfile
	1042
	1043	. /opt/modules/default/init/ksh
[493]	1044	$module_calls
[437]	1045
	1046	%%END%%
	1047
	1048	fi
	1049
[1]	1050	elif [[ $remote_host = nech ]]
	1051	then
	1052
	1053	if (( nodes > 1 ))
	1054	then
	1055	cat > $job_to_send << %%END%%
	1056	#!/bin/ksh
[622]	1057	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1058	#PBS -l ${qsubmem}=${Memory}gb
	1059	#PBS -b $nodes
	1060	#PBS -o $remote_dayfile
	1061	#PBS -N palm
	1062	#PBS -j o
	1063	#PBS -T mpisx
	1064
	1065	%%END%%
	1066
	1067	elif [[ $numprocs != 0 ]]
	1068	then
	1069	cat > $job_to_send << %%END%%
	1070	#!/bin/ksh
[622]	1071	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1072	#PBS -l ${qsubmem}=${Memory}gb
	1073	#PBS -o $remote_dayfile
	1074	#PBS -N palm
	1075	#PBS -j o
	1076
	1077	%%END%%
	1078
	1079	else
	1080	cat > $job_to_send << %%END%%
	1081	#!/bin/ksh
	1082	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1083	#PBS -o $remote_dayfile
	1084	#PBS -j o
	1085
	1086	%%END%%
	1087
	1088	fi
	1089
[251]	1090	elif [[ $remote_host = necriam ]]
[35]	1091	then
	1092
	1093	if (( nodes > 1 ))
	1094	then
	1095	cat > $job_to_send << %%END%%
	1096	#!/bin/ksh
	1097	#PBS -b $nodes
	1098	#PBS -o $remote_dayfile
[253]	1099	#PBS -N $job_name
[35]	1100	#PBS -j o
[251]	1101	#PBS -v MPIPROGINV=YES
[35]	1102
	1103	%%END%%
	1104
	1105	elif [[ $numprocs != 0 ]]
	1106	then
	1107	cat > $job_to_send << %%END%%
	1108	#!/bin/ksh
	1109	#PBS -q ${queue}
	1110	#PBS -o $remote_dayfile
[253]	1111	#PBS -N $job_name
[35]	1112	#PBS -j o
[251]	1113	#PBS -v MPIPROGINV=YES
[35]	1114
	1115	%%END%%
	1116
	1117	else
	1118	cat > $job_to_send << %%END%%
	1119	#!/bin/ksh
	1120	#PBS -o $remote_dayfile
	1121	#PBS -j o
	1122
	1123	%%END%%
	1124
	1125	fi
	1126
[1]	1127	elif [[ $remote_host = lctit ]]
	1128	then
	1129	cat > $job_to_send << %%END%%
[635]	1130	#!/bin/ksh
[1]	1131
[678]	1132	$module_calls
	1133
[1]	1134	%%END%%
	1135
	1136	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
[678]	1137	if [[ $tasks_per_node != $processes_per_node ]]
	1138	then
	1139	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -o $remote_dayfile -j oe -q $queue "
	1140	else
	1141	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -l place=scatter -o $remote_dayfile -j oe -q $queue "
	1142	fi
[1]	1143
	1144	else
	1145
	1146	cat > $job_to_send << %%END%%
	1147	# @\$-q ${queue}
	1148	# @\$-l${qsubtime} $timestring
	1149	# @\$-l${qsubmem} ${memory}mb
	1150	# @\$-o $remote_dayfile
	1151	# @\$-eo
	1152
	1153	%%END%%
	1154
	1155	fi
	1156
	1157
	1158	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	1159	# BEI EXIT VERANLASSEN
	1160	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	1161	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	1162	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1163	then
	1164	echo "set +vx" >> $job_to_send
	1165	echo "trap '" >> $job_to_send
	1166	echo "set +vx" >> $job_to_send
[635]	1167	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec ]]
[1]	1168	then
[622]	1169	if [[ $remote_host = ibmh ]]
[1]	1170	then
	1171	return_queue=c1
[693]	1172	elif [[ $remote_host = ibmkisti ]]
	1173	then
	1174	return_queue=class.1-2
[622]	1175	elif [[ $remote_host = ibmku ]]
	1176	then
	1177	return_queue=sdbg2
[1]	1178	elif [[ $remote_host = ibms ]]
	1179	then
	1180	return_queue=p_normal
	1181	elif [[ $remote_host = ibmy ]]
	1182	then
	1183	return_queue=serial
[325]	1184	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
	1185	then
[326]	1186	return_queue=serialq
[251]	1187	elif [[ $remote_host = necriam ]]
[1]	1188	then
[253]	1189	return_queue=SP
[1]	1190	else
	1191	return_queue=unknown
	1192	fi
	1193
	1194	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1195	then
	1196
[622]	1197	if [[ $remote_host = ibmku ]]
	1198	then
	1199	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1200	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1201	else
	1202	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1203	fi
[1]	1204	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	1205	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	1206	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	1207	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	1208	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	1209	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
[312]	1210	if [[ $host != "ibmh" ]]
	1211	then
	1212	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	1213	fi
[1]	1214	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	1215	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	1216
	1217	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	1218	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	1219
	1220	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	1221	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
[622]	1222	if [[ $remote_host = ibmku ]]
	1223	then
	1224	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
	1225	fi
[1]	1226	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	1227
	1228	elif [[ $remote_host = nech ]]
	1229	then
	1230	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1231	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1232	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1233	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1234	echo "#PBS -j o" >> $job_to_send
	1235	echo " " >> $job_to_send
	1236	echo "set -x" >> $job_to_send
	1237	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1238	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1239	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1240	echo "%%END%%" >> $job_to_send
	1241
[251]	1242	elif [[ $remote_host = necriam ]]
[35]	1243	then
	1244	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1245	echo "#PBS -q $return_queue" >> $job_to_send
	1246	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1247	echo "#PBS -j o" >> $job_to_send
	1248	echo " " >> $job_to_send
	1249	echo "set -x" >> $job_to_send
	1250	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1251	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1252	echo "%%END%%" >> $job_to_send
	1253
[181]	1254	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1255	then
	1256	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1257	echo "#!/bin/bash" >> $job_to_send
	1258	echo "#PBS -N job_protocol_transfer" >> $job_to_send
[325]	1259	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[181]	1260	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[366]	1261	echo "#PBS -l feature=data" >> $job_to_send
[181]	1262	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1263	echo "#PBS -j oe" >> $job_to_send
	1264	echo " " >> $job_to_send
	1265	echo ". /usr/share/modules/init/bash" >> $job_to_send
	1266	echo "set -x" >> $job_to_send
	1267	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1268	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1269	echo "%%END%%" >> $job_to_send
	1270
[1]	1271	else
	1272
	1273	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1274	echo "# @\\\$-q $return_queue" >> $job_to_send
	1275	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1276	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1277	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1278	then
	1279	echo "# @\$-l mpp_p=0" >> $job_to_send
	1280	fi
	1281	echo '# @\$-lF 10mb' >> $job_to_send
	1282	# echo '# @\$-o /dev/null' >> $job_to_send
	1283	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1284	echo '# @\\\$-eo' >> $job_to_send
	1285	echo " " >> $job_to_send
	1286	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1287	then
	1288	echo "set +vx" >> $job_to_send
	1289	echo ". .profile" >> $job_to_send
	1290	fi
	1291	echo "set -x" >> $job_to_send
	1292	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	1293	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1294	echo "%%END%%" >> $job_to_send
	1295	fi
	1296
	1297	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1298	then
	1299	echo "llsubmit scpjob.$kennung" >> $job_to_send
[181]	1300	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1301	then
	1302	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1303	echo "msub scpjob.$kennung" >> $job_to_send
[1]	1304	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1305	then
	1306	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	1307	elif [[ $remote_host = t3es ]]
	1308	then
	1309	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
	1310	else
	1311	echo "qsub scpjob.$kennung" >> $job_to_send
	1312	fi
[622]	1313	if [[ $remote_host != ibmku ]]
	1314	then
	1315	echo "rm scpjob.$kennung" >> $job_to_send
	1316	fi
[1]	1317	if [[ $remote_host = nech ]]
	1318	then
	1319	echo "cd -" >> $job_to_send
	1320	fi
	1321	else
	1322	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1323	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1324	fi
	1325	echo "set -x" >> $job_to_send
	1326	echo " ' exit" >> $job_to_send
	1327	echo "set -x" >> $job_to_send
	1328	fi
	1329
	1330
	1331
	1332	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	1333	cat $file_to_send >> $job_to_send
[69]	1334
[1]	1335	if [[ $remote_host = ibm ]]
	1336	then
	1337	echo " " >> $job_to_send
	1338	echo "exit" >> $job_to_send
	1339	fi
[635]	1340
	1341	# remove job file
[622]	1342	if [[ $remote_host = lctit \|\| $remote_host = ibmku ]]
[69]	1343	then
	1344	echo " " >> $job_to_send
	1345	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1346	fi
[1]	1347
	1348
	1349
	1350	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	1351	if [[ -z $remote_user ]]
	1352	then
	1353	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	1354	then
	1355	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	1356	fi
	1357	fi
	1358
	1359
	1360
	1361	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	1362	if [[ $no_submit = false ]]
	1363	then
	1364	if [[ $remote_host != $local_host ]]
	1365	then
	1366	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	1367	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	1368	then
[82]	1369	job_catalog_save=$job_catalog
	1370	job_catalog=job_queue
	1371	elif [[ $remote_host = nech ]]
	1372	then
	1373	job_catalog_save=$job_catalog
	1374	job_catalog=/hpf/b/${remote_user}/job_queue
	1375	fi
	1376	if [[ $local_host = decalpha ]]
	1377	then
	1378	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	1379	# PASSWORT AB
	1380	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	1381	elif [[ $remote_host = nech ]]
	1382	then
	1383	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	1384	# TRANSFERIERT WERDEN
	1385	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1386	else
[82]	1387	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1388	fi
[82]	1389	if [[ $? = 1 ]]
	1390	then
	1391	locat=scp; exit
	1392	fi
	1393	if [[ $remote_host = ibms ]]
	1394	then
	1395	job_catalog=$job_catalog_save
	1396	fi
[1]	1397	[[ $verify = true ]] && printf "\n >>> finished\n"
	1398	else
	1399	eval job_catalog=$job_catalog
	1400	cp $job_to_send ${job_catalog}/$job_on_remhost
	1401	fi
	1402
	1403
	1404
	1405	# NQS- BZW. LOADLEVELER-JOB STARTEN
	1406	if [[ $remote_host != $local_host ]]
	1407	then
	1408	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1409
	1410	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
[1]	1411	then
[635]	1412	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
	1413	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
	1414	elif [[ $remote_host = ibmku ]]
	1415	then
	1416	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1]	1417	else
[635]	1418	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1419	fi
[1]	1420
	1421	[[ $verify = true ]] && printf " >>> o.k.\n"
	1422	else
	1423	cd $job_catalog
[210]	1424	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
[1]	1425	then
	1426	eval $submcom $job_on_remhost
[717]	1427	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 \|\| $localhost = lck \|\| $localhost = lckordi\|\| $localhost = lcyon ]]
[108]	1428	then
[635]	1429	chmod u+x $job_on_remhost
	1430	echo "$submcom $job_on_remhost"
[108]	1431	eval $submcom $job_on_remhost
[1]	1432	elif [[ $local_host = nech ]]
	1433	then
	1434	if [[ $queue = default ]]
	1435	then
	1436	eval $submcom $job_on_remhost
	1437	else
	1438	eval $submcom -q $queue $job_on_remhost
	1439	fi
	1440	else
	1441	qsub $job_on_remhost
	1442	fi
[622]	1443
	1444	# Jobfile must not be deleted on lctit/ibmku!! This will be done
	1445	# only at the end of the job.
	1446	if [[ $local_host != lctit && $local_host != ibmku ]]
	1447	then
	1448	rm $job_on_remhost
	1449	fi
[1]	1450	cd - > /dev/null
	1451	fi
	1452	fi
	1453
	1454
	1455
	1456	# ABSCHLUSSARBEITEN
	1457
	1458	if [[ $no_submit = false ]]
	1459	then
	1460	rm $job_to_send
	1461	fi
	1462	[[ $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 |