Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1044

Last change on this file since 1044 was 1043, checked in by raasch, 12 years ago
further script adjustments for lckiaps
Property svn:keywords set to `Id Rev`
File size: 52.0 KB

Rev	Line
[1]	1	#!/bin/ksh
[82]	2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
[169]	3	# $Id: subjob 1043 2012-11-06 06:18:09Z maronga $
[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
[739]	132	# 17/08/11 - Siggi - hicegate0 added
[748]	133	# 18/08/11 - Siggi - workaround on lcsgi in order to avoid appends to file
	134	# last_job_transfer_protocol
[749]	135	# 21/08/11 - Siggi - inferno admitted
[789]	136	# 29/11/11 - Siggi - adjustments for lcsgih/lcsgib queues, pingui admitted
[800]	137	# 21/12/11 - Theres- solano admitted
[892]	138	# 31/01/12 - Matthias - adjustments for ibmh
	139	# 02/05/12 - Bjoern - lcxt4 is now lcxe6, adjustments for lcxe6,
	140	# bugfixes for the parameter file check, initial
	141	# commands on ssh calls sourced out to .mrun.config
[1021]	142	# 02/10/12 - Siggi - adjustments for lcsb
[1040]	143	# 01/11/12 - Siggi - adjustments for lckiaps
[1]	144
[437]	145
[1021]	146
[1]	147	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
[352]	148	delete_dayfile=false
[799]	149	email_notification=none
[122]	150	group_number=none
[1]	151	locat=normal
	152	no_default_queue=none
	153	no_submit=false
	154	job_catalog="~/job_queue"
	155	job_name=none
	156	local_user=$LOGNAME
	157	node_usage=shared
[475]	158	numprocs=0
[1]	159	punkte="..........................................................."
	160	submcom=qsub
	161	queue=default
	162	remote_host=none
	163	remote_user=""
	164	verify=true
	165
	166	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	167	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	168	typeset -L20 spalte1
	169	typeset -R40 spalte2
	170	typeset -L60 spalte3
	171
	172
	173
	174	# FEHLERBEHANDLUNG
	175	# BEI EXIT:
	176	trap 'if [[ $locat != normal ]]
	177	then
	178	case $locat in
	179	(option) printf "\n --> available optios can be displayed"
	180	printf " by typing:"
	181	printf "\n \"subjob ?\" \n";;
	182	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	183	(*) printf "\n +++ unknown error"
	184	printf "\n please inform S. Raasch!\n"
	185	esac
	186	[[ -f $job_to_send ]] && rm $job_to_send
	187	printf "\n\n+++ SUBJOB killed \n\n"
	188	fi' exit
	189
	190
	191	# BEI TERMINAL-BREAK:
	192	trap '[[ -f $job_to_send ]] && rm $job_to_send
	193	printf "\n\n+++ SUBJOB killed \n\n"
	194	exit
	195	' 2
	196
	197
	198
	199
	200	# LOKALEN HOSTNAMEN FESTSTELLEN
	201	local_host=$(hostname)
	202
	203
	204
	205	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
	206	# UEBERHAUPT ZULAESSIG IST
[717]	207	# Note: One of the entries for "lck" or "lckordi" always has to be
	208	# comment out, because the hostname (node*) is same for both machines
[1]	209	case $local_host in
	210	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[102]	211	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	212	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[400]	213	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
	214	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
	215	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
	216	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
	217	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
[291]	218	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[749]	219	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[12]	220	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[372]	221	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[693]	222	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
[108]	223	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
[1]	224	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	225	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[108]	226	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
[367]	227	(node*) local_addres=165.132.26.61 local_host=lck;;
[717]	228	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
[693]	229	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
[101]	230	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[1]	231	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	232	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	233	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	234	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[892]	235	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
[400]	236	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
	237	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
	238	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
	239	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
[739]	240	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
[292]	241	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
[210]	242	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
[440]	243	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
[749]	244	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
[5]	245	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
[407]	246	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
[5]	247	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
[1040]	248	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
[5]	249	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[437]	250	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
[892]	251	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
[1]	252	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	253	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	254	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	255	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[789]	256	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
[5]	257	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[622]	258	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
[1]	259	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
[799]	260	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
[1]	261	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
[259]	262	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
[181]	263	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
[196]	264	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
[635]	265	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
[1021]	266	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
[1]	267	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
[622]	268	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
[1]	269	(*) printf "\n +++ \"$local_host\" unknown";
	270	printf "\n please inform S. Raasch!";
	271	locat=parameter; exit;;
	272	esac
	273
	274
	275
	276	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	277	remote_host=$local_host
	278
	279
	280
	281
	282	# PROZEDUROPTIONEN EINLESEN
[352]	283	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
[1]	284	do
	285	case $option in
	286	(c) job_catalog=$OPTARG;;
	287	(d) delete_dayfile=true;;
	288	(D) no_submit=true;;
[352]	289	(e) email_notification=$OPTARG;;
[125]	290	(g) group_number=$OPTARG;;
[1]	291	(h) remote_host=$OPTARG;;
	292	(m) memory=$OPTARG;;
	293	(n) job_name=$OPTARG;;
	294	(N) node_usage=$OPTARG;;
	295	(O) threads_per_task=$OPTARG;;
	296	(q) no_default_queue=$OPTARG;;
	297	(t) cputime=$OPTARG;;
	298	(T) tasks_per_node=$OPTARG;;
	299	(u) remote_user=$OPTARG;;
	300	(v) verify=false;;
	301	(X) numprocs=$OPTARG;;
	302	(\?) printf "\n +++ Option $OPTARG unknown \n";
	303	locat=option; exit;;
	304	esac
	305	done
	306
	307
	308	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	309	shift OPTIND-1; file_to_send=$1
	310
	311
	312	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	313	if [ "$1" = "?" ]
	314	then
	315	(printf "\n *** subjob can be called as follows:\n"
	316	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	317	printf "\n Description of available options:\n"
	318	printf "\n Option Description Default-Value"
	319	printf "\n -c job-input- and output-catalog ~/job_queue"
	320	printf "\n -d no job-protocol will be created ---"
	321	printf "\n -D only the job-file will be created ---"
	322	printf "\n -h execution host, available hosts: $remote_host"
[693]	323	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
[1040]	324	printf "\n lckiaps, lctit, nech, necriam, unics"
[1]	325	printf "\n -m memory demand per process in MByte ---"
	326	printf "\n -n jobname <jobdatei>"
	327	printf "\n -O threads per task (for OpenMP usage) 1"
	328	printf "\n -q job-queue to be used default"
	329	printf "\n -t allowed cpu-time in seconds ---"
	330	printf "\n -T tasks per node (on parallel hosts) ---"
	331	printf "\n -u username on execution host from .netrc"
	332	printf "\n -v no prompt for confirmation ---"
	333	printf "\n -X # of processors (on parallel hosts) 1"
	334	printf "\n "
	335	printf "\n The only possible positional parameter is <jobfile>:"
	336	printf "\n The complete NQS-job must be provided here."
	337	printf "\n <jobfile>=? creates this outline\n\n") \| more
	338	exit
	339	fi
	340
	341
	342
	343	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	344	if [[ "$file_to_send" = "" ]]
	345	then
	346	printf "\n +++ job-file missing"
	347	locat=parameter; exit
	348	else
	349	if [[ -f $file_to_send ]]
	350	then
	351	true
	352	else
	353	printf "\n +++ job-file: "
	354	printf "\n $file_to_send"
	355	printf "\n does not exist"
	356	locat=parameter; exit
	357	fi
	358	fi
	359
	360
	361
	362	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	363	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	364	if [[ $job_name = none ]]
	365	then
	366	job_name=$file_to_send
	367	fi
	368	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	369	then
	370	printf "\n +++ job-file name: "
	371	printf "\n $job_name"
	372	printf "\n must not contain \"/\"-characters"
	373	locat=parameter; exit
	374	fi
	375
	376
	377
	378
	379	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	380	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	381	if [[ $remote_host = none ]]
	382	then
	383	printf "\n +++ host missing"
	384	locat=option; exit
	385	else
	386	case $remote_host in
[251]	387	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[312]	388	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[693]	389	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[622]	390	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
[251]	391	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	392	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	393	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
[440]	394	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
[540]	395	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
	396	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
[367]	397	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[1040]	398	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
[717]	399	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
[1021]	400	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
[635]	401	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
[892]	402	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
[437]	403	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
[693]	404	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
[251]	405	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	406	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
	407	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	408	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	409	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	410	locat=parameter; exit;;
[1]	411	esac
	412	fi
	413
	414
	415	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	416	if [[ $no_default_queue != none ]]
	417	then
	418	error=false
	419	ndq=$no_default_queue
	420	case $remote_host in
	421	(ibm) case $ndq in
	422	(p690_express\|p690_standard\|p690_long) error=false;;
	423	(*) error=true;;
	424	esac;;
	425	(ibmh) case $ndq in
[312]	426	(no_class) error=false;;
[1]	427	(*) error=true;;
	428	esac;;
[693]	429	(ibmkisti) case $ndq in
	430	(class.32plus\|class.1-2\|class.2-32) error=false;;
	431	(*) error=true;;
	432	esac;;
[622]	433	(ibmku) case $ndq in
	434	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	435	(*) error=true;;
	436	esac;;
[1]	437	(ibms) case $ndq in
	438	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	439	(*) error=true;;
	440	esac;;
	441	(ibmy) case $ndq in
	442	(parallel) error=false;;
	443	(*) error=true;;
	444	esac;;
[1040]	445	(lckiaps) case $ndq in
	446	(express\|normal) error=false;;
	447	(*) error=true;;
	448	esac;;
[440]	449	(lckyoto) case $ndq in
	450	(eh\|ph) error=false;;
	451	(*) error=true;;
	452	esac;;
[191]	453	(lcsgib\|lcsgih) case $ndq in
[416]	454	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
[191]	455	(*) error=true;;
	456	esac;;
[1]	457	(lctit) case $ndq in
[635]	458	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	459	(*) error=true;;
	460	esac;;
	461	(t3eb) case $ndq in
	462	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	463	(*) error=true;;
	464	esac;;
[251]	465	(necriam) case $ndq in
[253]	466	(SP\|SS\|P6) error=false;;
[251]	467	(*) error=true;;
	468	esac;;
[1]	469	(t3eh) case $ndq in
	470	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	471	(*) error=true;;
	472	esac;;
	473	(t3ej2\|t3ej5) case $ndq in
	474	(low\|normal\|high) error=false;;
	475	(*) error=true;;
	476	esac;;
	477	(t3es) case $ndq in
	478	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	479	(*) error=true;;
	480	esac;;
	481	(unics) case $ndq in
	482	(unics\|ht) error=false;;
	483	(*) error=true;;
	484	esac;;
	485	esac
	486	if [[ $error = true ]]
	487	then
	488	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	489	locat=parameter; exit
	490	else
	491	queue=$no_default_queue
	492	fi
	493	fi
	494
	495
	496
	497	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	498	# AUFTEILEN
	499	done=false
	500	while [[ $done = false ]]
	501	do
	502	if (( $cputime <= 0 ))
	503	then
	504	printf "\n +++ wrong cpu-time or cpu-time missing"
	505	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	506	printf "\n >>> "
	507	read cputime 1>/dev/null 2>&1
	508	else
	509	done=true
	510	fi
	511	done
	512	if [[ $remote_host = nech ]]
	513	then
	514	if (( tasks_per_node != 0 ))
	515	then
	516	(( cputime = cputime * tasks_per_node ))
	517	elif [[ $numprocs != 0 ]]
	518	then
	519	(( cputime = cputime * numprocs ))
	520	fi
	521	fi
	522	(( stunden = cputime / 3600 ))
	523	(( resttime = cputime - stunden * 3600 ))
	524	(( minuten = resttime / 60 ))
	525	(( sekunden = resttime - minuten * 60 ))
	526	timestring=${stunden}:${minuten}:${sekunden}
	527
	528
	529
	530	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	531	done=false
	532	while [[ $done = false ]]
	533	do
	534	if (( memory <= 0 ))
	535	then
	536	printf "\n +++ wrong memory demand or memory demand missing"
	537	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	538	printf "\n >>> "
	539	read memory 1>/dev/null 2>&1
	540	else
	541	done=true
	542	fi
	543	done
	544
[251]	545	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
[1]	546	then
	547	if (( tasks_per_node != 0 ))
	548	then
	549	(( Memory = memory * tasks_per_node / 1000 ))
	550	elif [[ $numprocs != 0 ]]
	551	then
	552	(( Memory = memory * numprocs / 1000 ))
	553	else
	554	(( Memory = memory / 1000 ))
	555	fi
[635]	556	elif [[ $remote_host = lctit ]]
	557	then
	558	(( Memory = memory * tasks_per_node / 1000 ))
[1]	559	fi
	560
	561
	562	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	563	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	564	then
	565	(( memory = memory * threads_per_task ))
	566	fi
	567
	568
	569	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	570	if (( tasks_per_node != 0 ))
	571	then
[622]	572	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
[1]	573	fi
	574
[622]	575	# Calculate number of processes per node
	576	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	577
[696]	578	# Calculate number of MPI tasks
	579	(( mpi_tasks = numprocs / threads_per_task ))
[1]	580
[696]	581
[1]	582	# HEADER-AUSGABE
	583	if [[ $verify = true ]]
	584	then
	585	printf "\n\n"
	586	printf "#--------------------------------------------------------------# \n"
	587	spalte1=SUBJOB;spalte2=$(date)
	588	printf "\| $spalte1$spalte2 \| \n"
	589	printf "\| \| \n"
	590	printf "\| values of parameters/options: \| \n"
	591	spalte1=local_host$punkte; spalte2=$punkte$local_host
	592	printf "\| $spalte1$spalte2 \| \n"
	593	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	594	printf "\| $spalte1$spalte2 \| \n"
	595	spalte1=queue$punkte; spalte2=$punkte$queue
	596	printf "\| $spalte1$spalte2 \| \n"
	597	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	598	printf "\| $spalte1$spalte2 \| \n"
	599	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	600	printf "\| $spalte1$spalte2 \| \n"
	601	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	602	printf "\| $spalte1$spalte2 \| \n"
	603	printf "#--------------------------------------------------------------# \n\n"
	604
	605
	606	# KONTROLLABFRAGE, OB ALLES O.K.
	607	antwort="dummy"
	608	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	609	do
	610	read antwort?" >>> continue (y/n) ? "
	611	done
	612	if [[ $antwort = n \|\| $antwort = N ]]
	613	then
	614	locat=verify; exit
	615	fi
	616	printf "\n"
	617	fi
	618
	619
	620
	621	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	622	kennung=$RANDOM
	623	job_on_remhost=${job_name}_${kennung}_$local_host
	624	job_to_send=job_to_send_$kennung
	625	if [[ $delete_dayfile = false ]]
	626	then
	627	remote_dayfile=${local_host}_${job_name}_result_$kennung
	628	local_dayfile=${remote_host}_${job_name}
	629	else
	630	remote_dayfile=/dev/null
	631	fi
	632
	633
	634
[622]	635	# Generate the batch job scripts (qsub/msub/LoadLeveler)
[1]	636	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	637	then
	638
[622]	639	# General LoadLeveler settings
	640	execute_in_shell="#!/bin/ksh"
	641	use_shell="# @ shell = /bin/ksh"
	642	consumable_memory="ConsumableMemory($memory mb)"
	643	class="# @ class = $queue"
	644	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	645	network_to_use="# @ network.mpi = sn_all,shared,us"
	646	data_limit="# @ data_limit = 1.76gb"
	647	image_size="# @ image_size = 50"
[693]	648	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	649
[693]	650	if [[ $email_notification = none ]]
	651	then
	652	notify_user=""
	653	else
	654	notify_user="# @ notify_user = $email_notification"
	655	if [[ $delete_dayfile = true ]]
	656	then
	657	notification='# @ notification = never'
	658	fi
	659	fi
[622]	660
[312]	661	if [[ $remote_host = ibmh ]]
[1]	662	then
[312]	663	data_limit=""
	664	network_to_use=""
	665	class=""
	666	environment=""
[814]	667	rset="# @ rset = RSET_MCM_AFFINITY"
	668	task_affinity="# @ task_affinity = core(1)"
[693]	669	elif [[ $remote_host = ibmkisti ]]
	670	then
	671	network_to_use="# @ network.MPI = sn_all,shared,US"
	672	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	673	if [[ $threads_per_task = 1 ]]
	674	then
	675	rset="# @ rset = RSET_MCM_AFFINITY"
	676	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	677	fi
[693]	678	environment=""
	679	use_shell=""
	680	data_limit=""
	681	image_size=""
[622]	682	elif [[ $remote_host = ibmku ]]
	683	then
	684	execute_in_shell="#!/usr/bin/ksh"
	685	use_shell="# @ shell = /usr/bin/ksh"
	686	consumable_memory=""
	687	environment=""
	688	network_to_use="# @ network.mpi = sn_all,shared,us"
	689	data_limit=""
	690	image_size=""
	691	elif [[ $remote_host = ibms ]]
	692	then
	693	network_to_use="# @ network.mpi = csss,shared,us"
	694	elif [[ $remote_host = ibmy ]]
	695	then
	696	consumable_memory=""
	697	network_to_use=""
[1]	698	fi
	699
	700	cat > $job_to_send << %%END%%
[622]	701	$execute_in_shell
	702	$use_shell
[1]	703
	704	# @ job_type = parallel
[693]	705	# @ job_name = $job_name
[1]	706	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	707	# @ output = $remote_dayfile
	708	# @ error = $remote_dayfile
[693]	709	$wall_clock_limit
[622]	710	$image_size
[312]	711	$class
	712	$environment
[1]	713	$network_to_use
	714	$data_limit
[693]	715	$rset
	716	$mcm_affinity_options
[814]	717	$task_affinity
[1]	718	$notification
[693]	719	$notify_user
[1]	720
	721	%%END%%
	722
	723	if (( nodes > 0 ))
	724	then
	725
[693]	726	if [[ $remote_host != ibmkisti ]]
	727	then
	728
	729	cat >> $job_to_send << %%END%%
[1]	730	# @ node = $nodes
[622]	731	# @ tasks_per_node = $processes_per_node
[1]	732	# @ node_usage = $node_usage
	733	# @ queue
	734
	735	%%END%%
	736
[693]	737	else
	738
	739	cat >> $job_to_send << %%END%%
[696]	740	# @ total_tasks = $mpi_tasks
[693]	741	# @ blocking = unlimited
	742	# @ queue
	743
	744	%%END%%
	745
	746	fi
	747
[1]	748	else
	749
	750	if [[ $remote_host != ibmy ]]
	751	then
	752
	753	cat >> $job_to_send << %%END%%
	754	# @ blocking = unlimited
	755	# @ total_tasks = $numprocs
	756	# @ node_usage = $node_usage
	757	# @ queue
	758
	759	%%END%%
	760
	761	else
	762
	763	cat >> $job_to_send << %%END%%
	764	# @ node = 1
	765	# @ total_tasks = $numprocs
	766	# @ queue
	767
	768	%%END%%
	769
	770	fi
	771
	772	fi
	773
[696]	774	# workaround because of silly job filter on ibmkisti
	775	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	776	then
	777	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	778	fi
	779
[1]	780	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	781	then
	782
	783	cat > $job_to_send << %%END%%
	784	#!/bin/ksh
	785
	786	# @ job_type = serial
	787	# @ node_usage = $node_usage
	788	# @ job_name = palm
	789	# @ wall_clock_limit = ${timestring},$timestring
	790	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	791	# @ output = $remote_dayfile
	792	# @ error = $remote_dayfile
[312]	793	$class
[1]	794	$notification
	795
	796	# @ queue
	797
	798	%%END%%
	799
[108]	800	elif [[ $remote_host = lcfimm ]]
	801	then
	802
	803	if [[ $numprocs != 0 ]]
	804	then
	805	cat > $job_to_send << %%END%%
	806	#!/bin/ksh
	807	#PBS -N $job_name
[552]	808	#PBS -A $project_account
[108]	809	#PBS -l walltime=$timestring
[622]	810	#PBS -l nodes=${nodes}:ppn=$processes_per_node
[108]	811	#PBS -l pmem=${memory}mb
	812	#PBS -m abe
	813	#PBS -o $remote_dayfile
	814	#PBS -j oe
	815	mpd &
	816
	817	%%END%%
	818
[168]	819	else
	820	cat > $job_to_send << %%END%%
	821	#!/bin/ksh
	822	#PBS -N $job_name
[552]	823	#PBS -A $project_account
[168]	824	#PBS -l walltime=$timestring
	825	#PBS -l ncpus=1
	826	#PBS -l pmem=${memory}mb
	827	#PBS -m abe
	828	#PBS -o $remote_dayfile
	829	#PBS -j oe
	830
	831	%%END%%
	832
	833	fi
	834
[1021]	835	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
[368]	836	then
	837
	838	if [[ $numprocs != 0 ]]
	839	then
	840	cat > $job_to_send << %%END%%
	841	#!/bin/ksh
	842	#PBS -N $job_name
	843	#PBS -l walltime=$timestring
	844	#PBS -l ncpus=$numprocs
	845	#PBS -l pmem=${memory}mb
	846	#PBS -o $remote_dayfile
[1021]	847	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[368]	848	#PBS -j oe
	849
	850	mpd &
	851
	852	%%END%%
	853
	854	else
	855	cat > $job_to_send << %%END%%
	856	#!/bin/ksh
	857	#PBS -N $job_name
	858	#PBS -l walltime=$timestring
	859	#PBS -l ncpus=1
	860	#PBS -l pmem=${memory}mb
	861	#PBS -o $remote_dayfile
	862	#PBS -j oe
	863
	864	%%END%%
	865
	866	fi
	867
[1040]	868	elif [[ $remote_host = lckiaps ]]
	869	then
	870
	871	if [[ $numprocs != 0 ]]
	872	then
	873	cat > $job_to_send << %%END%%
	874	#!/bin/ksh
	875	#PBS -N $job_name
	876	#PBS -l walltime=$timestring
	877	#PBS -l select=1:ncpus=$numprocs
	878	#PBS -l pmem=${memory}mb
	879	#PBS -q $queue
	880	#PBS -o $remote_dayfile
	881	#PBS -j oe
	882	#PBS -V
	883
	884	%%END%%
	885
	886	else
	887	cat > $job_to_send << %%END%%
	888	#!/bin/ksh
	889	#PBS -N $job_name
	890	#PBS -l walltime=$timestring
	891	#PBS -l ncpus=1
	892	#PBS -l pmem=${memory}mb
	893	#PBS -o $remote_dayfile
	894	#PBS -j oe
	895
	896	%%END%%
	897
	898	fi
	899
[693]	900	elif [[ $remote_host = lcyon ]]
	901	then
	902
	903	if [[ $numprocs != 0 ]]
	904	then
	905	cat > $job_to_send << %%END%%
	906	#!/bin/ksh
	907	#PBS -N $job_name
	908	#PBS -l walltime=$timestring
	909	#PBS -l ncpus=$numprocs
	910	#PBS -l pmem=${memory}mb
	911	#PBS -o $remote_dayfile
	912	#PBS -j oe
	913
	914	%%END%%
	915
	916	else
	917	cat > $job_to_send << %%END%%
	918	#!/bin/ksh
	919	#PBS -N $job_name
	920	#PBS -l walltime=$timestring
	921	#PBS -l ncpus=1
	922	#PBS -l pmem=${memory}mb
	923	#PBS -o $remote_dayfile
	924	#PBS -j oe
	925
	926	%%END%%
	927
	928	fi
	929
[181]	930	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
[179]	931	then
	932
[376]	933	if [[ "$sgi_feature" != "" ]]
	934	then
	935	feature_directive="#PBS -l feature=$sgi_feature"
	936	else
	937	feature_directive=""
	938	fi
[799]	939
[416]	940	if [[ $queue = dataq \|\| $queue = permq ]]
[399]	941	then
	942	feature_directive="#PBS -l feature=data"
	943	fi
[376]	944
[789]	945	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
[191]	946	then
	947	queue_directive="#PBS -q $queue"
	948	else
	949	queue_directive=""
	950	fi
	951
[352]	952	if [[ $email_notification = none ]]
	953	then
	954	email_directive=""
	955	else
	956	email_directive="#PBS -M $email_notification"
	957	fi
	958
[179]	959	if [[ $numprocs != 0 ]]
	960	then
	961	cat > $job_to_send << %%END%%
[181]	962	#!/bin/bash
[179]	963	#PBS -N $job_name
	964	#PBS -l walltime=$timestring
[622]	965	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[181]	966	#PBS -l naccesspolicy=$node_usage
[179]	967	#PBS -o $remote_dayfile
	968	#PBS -j oe
[376]	969	$feature_directive
[191]	970	$queue_directive
[352]	971	$email_directive
[179]	972
[622]	973	eval \`/sw/swdist/bin/modulesinit\`
	974	#. /usr/share/modules/init/bash
[892]	975	$init_cmds
[493]	976	$module_calls
[374]	977
[181]	978	echo ld_library_path=\$LD_LIBRARY_PATH
[179]	979
	980	%%END%%
	981
	982	else
	983	cat > $job_to_send << %%END%%
	984	#PBS -S /bin/bash
	985	#PBS -N $job_name
	986	#PBS -l walltime=$timestring
	987	#PBS -l ncpus=1
	988	#PBS -o $remote_dayfile
[475]	989	#PBS -j oe
[400]	990	$queue_directive
	991	$email_directive
[179]	992
[622]	993	eval \`/sw/swdist/bin/modulesinit\`
	994	#. /usr/share/modules/init/bash
[892]	995	$init_cmds
[493]	996	$module_calls
[475]	997
[179]	998	%%END%%
	999
	1000	fi
	1001
[892]	1002	elif [[ $remote_host = lcxe6 ]]
[164]	1003	then
	1004
	1005	if [[ $numprocs != 0 ]]
	1006	then
	1007	cat > $job_to_send << %%END%%
[799]	1008	#!/bin/ksh
[164]	1009	#PBS -S /bin/ksh
	1010	#PBS -N $job_name
[552]	1011	#PBS -A $project_account
[206]	1012	#PBS -j oe
[164]	1013	#PBS -l walltime=$timestring
	1014	#PBS -l mppwidth=${numprocs}
[622]	1015	#PBS -l mppnppn=${processes_per_node}
[164]	1016	#PBS -m abe
	1017	#PBS -o $remote_dayfile
[492]	1018	$email_directive
[164]	1019
[892]	1020	$init_cmds
[493]	1021	$module_calls
[343]	1022
[164]	1023	%%END%%
	1024
[108]	1025	else
	1026	cat > $job_to_send << %%END%%
	1027	#!/bin/ksh
[168]	1028	#PBS -S /bin/ksh
[108]	1029	#PBS -N $job_name
[552]	1030	#PBS -A $project_account
[206]	1031	#PBS -j oe
[108]	1032	#PBS -l walltime=$timestring
	1033	#PBS -l ncpus=1
	1034	#PBS -l pmem=${memory}mb
	1035	#PBS -m abe
[492]	1036	$email_directive
[108]	1037	#PBS -o $remote_dayfile
	1038
[892]	1039	$init_cmds
[493]	1040	$module_calls
[343]	1041
[108]	1042	%%END%%
	1043
	1044	fi
	1045
[440]	1046	elif [[ $remote_host = lckyoto ]]
[437]	1047	then
	1048
[440]	1049	cat > $job_to_send << %%END%%
[799]	1050	#!/bin/ksh
[440]	1051	# @\$-o $remote_dayfile
	1052	# @\$-eo -oi
	1053	# @\$-lP 16
[799]	1054	# @\$-lp 1
[440]	1055	# @\$-lm 28gb -llm unlimited -ls unlimited
	1056	# @\$-q $queue
	1057	# @\$-Pvn abs_pack
	1058	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	1059	#. /thin/local/etc/setprofile/intel-11.0.sh
	1060	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1061	. ~/.myprofile
	1062	#. /home2/t/t51254/palm/current_version/myprofile
	1063	#. /thin/apps/pgi/mpi.sh
	1064	#
	1065	env
	1066	#
	1067	set -x
	1068
	1069	%%END%%
	1070
	1071	elif [[ $remote_host = lcxt5m ]]
	1072	then
	1073
[437]	1074	if [[ $numprocs != 0 ]]
	1075	then
	1076	cat > $job_to_send << %%END%%
[799]	1077	#!/bin/ksh
[437]	1078	#PBS -S /bin/ksh
	1079	#PBS -N $job_name
	1080	#PBS -j oe
	1081	#PBS -l walltime=$timestring
	1082	#PBS -l mppwidth=${numprocs}
[622]	1083	#PBS -l mppnppn=${processes_per_node}
[437]	1084	#PBS -m abe
	1085	#PBS -o $remote_dayfile
	1086
[892]	1087	$init_cmds
[493]	1088	$module_calls
[437]	1089
	1090	%%END%%
	1091
	1092	else
	1093	cat > $job_to_send << %%END%%
	1094	#!/bin/ksh
	1095	#PBS -S /bin/ksh
	1096	#PBS -N $job_name
	1097	#PBS -j oe
	1098	#PBS -l walltime=$timestring
	1099	#PBS -l ncpus=1
	1100	#PBS -l pmem=${memory}mb
	1101	#PBS -m abe
	1102	#PBS -o $remote_dayfile
	1103
[892]	1104	$init_cmds
[493]	1105	$module_calls
[437]	1106
	1107	%%END%%
	1108
	1109	fi
	1110
[1]	1111	elif [[ $remote_host = nech ]]
	1112	then
	1113
	1114	if (( nodes > 1 ))
	1115	then
	1116	cat > $job_to_send << %%END%%
	1117	#!/bin/ksh
[622]	1118	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1119	#PBS -l ${qsubmem}=${Memory}gb
	1120	#PBS -b $nodes
	1121	#PBS -o $remote_dayfile
	1122	#PBS -N palm
	1123	#PBS -j o
	1124	#PBS -T mpisx
	1125
	1126	%%END%%
	1127
	1128	elif [[ $numprocs != 0 ]]
	1129	then
	1130	cat > $job_to_send << %%END%%
	1131	#!/bin/ksh
[622]	1132	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1133	#PBS -l ${qsubmem}=${Memory}gb
	1134	#PBS -o $remote_dayfile
	1135	#PBS -N palm
	1136	#PBS -j o
	1137
	1138	%%END%%
	1139
	1140	else
	1141	cat > $job_to_send << %%END%%
	1142	#!/bin/ksh
	1143	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1144	#PBS -o $remote_dayfile
	1145	#PBS -j o
	1146
	1147	%%END%%
	1148
	1149	fi
	1150
[251]	1151	elif [[ $remote_host = necriam ]]
[35]	1152	then
	1153
	1154	if (( nodes > 1 ))
	1155	then
	1156	cat > $job_to_send << %%END%%
	1157	#!/bin/ksh
	1158	#PBS -b $nodes
	1159	#PBS -o $remote_dayfile
[253]	1160	#PBS -N $job_name
[35]	1161	#PBS -j o
[251]	1162	#PBS -v MPIPROGINV=YES
[35]	1163
	1164	%%END%%
	1165
	1166	elif [[ $numprocs != 0 ]]
	1167	then
	1168	cat > $job_to_send << %%END%%
	1169	#!/bin/ksh
	1170	#PBS -q ${queue}
	1171	#PBS -o $remote_dayfile
[253]	1172	#PBS -N $job_name
[35]	1173	#PBS -j o
[251]	1174	#PBS -v MPIPROGINV=YES
[35]	1175
	1176	%%END%%
	1177
	1178	else
	1179	cat > $job_to_send << %%END%%
	1180	#!/bin/ksh
	1181	#PBS -o $remote_dayfile
	1182	#PBS -j o
	1183
	1184	%%END%%
	1185
	1186	fi
	1187
[1]	1188	elif [[ $remote_host = lctit ]]
	1189	then
	1190	cat > $job_to_send << %%END%%
[635]	1191	#!/bin/ksh
[892]	1192	$init_cmds
[678]	1193	$module_calls
	1194
[1]	1195	%%END%%
	1196
	1197	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
[678]	1198	if [[ $tasks_per_node != $processes_per_node ]]
	1199	then
	1200	submcom="$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 $queue "
	1201	else
	1202	submcom="$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 "
	1203	fi
[1]	1204
	1205	else
	1206
	1207	cat > $job_to_send << %%END%%
	1208	# @\$-q ${queue}
	1209	# @\$-l${qsubtime} $timestring
	1210	# @\$-l${qsubmem} ${memory}mb
	1211	# @\$-o $remote_dayfile
	1212	# @\$-eo
	1213
	1214	%%END%%
	1215
	1216	fi
	1217
	1218
	1219	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	1220	# BEI EXIT VERANLASSEN
	1221	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	1222	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	1223	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1224	then
	1225	echo "set +vx" >> $job_to_send
	1226	echo "trap '" >> $job_to_send
	1227	echo "set +vx" >> $job_to_send
[1040]	1228	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lckiaps ]]
[1]	1229	then
[622]	1230	if [[ $remote_host = ibmh ]]
[1]	1231	then
	1232	return_queue=c1
[693]	1233	elif [[ $remote_host = ibmkisti ]]
	1234	then
	1235	return_queue=class.1-2
[622]	1236	elif [[ $remote_host = ibmku ]]
	1237	then
	1238	return_queue=sdbg2
[1]	1239	elif [[ $remote_host = ibms ]]
	1240	then
	1241	return_queue=p_normal
	1242	elif [[ $remote_host = ibmy ]]
	1243	then
	1244	return_queue=serial
[325]	1245	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
	1246	then
[326]	1247	return_queue=serialq
[251]	1248	elif [[ $remote_host = necriam ]]
[1]	1249	then
[253]	1250	return_queue=SP
[1040]	1251	elif [[ $remote_host = lckiaps ]]
	1252	then
	1253	return_queue=express
[1]	1254	else
	1255	return_queue=unknown
	1256	fi
	1257
	1258	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1259	then
	1260
[622]	1261	if [[ $remote_host = ibmku ]]
	1262	then
	1263	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1264	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1265	else
	1266	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1267	fi
[1]	1268	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	1269	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	1270	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	1271	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	1272	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	1273	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
[312]	1274	if [[ $host != "ibmh" ]]
	1275	then
	1276	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	1277	fi
[1]	1278	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	1279	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	1280
	1281	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	1282	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	1283
	1284	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	1285	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]	1286	if [[ $remote_host = ibmku ]]
	1287	then
	1288	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
	1289	fi
[1]	1290	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	1291
	1292	elif [[ $remote_host = nech ]]
	1293	then
	1294	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1295	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1296	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1297	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1298	echo "#PBS -j o" >> $job_to_send
	1299	echo " " >> $job_to_send
	1300	echo "set -x" >> $job_to_send
	1301	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1302	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1303	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1304	echo "%%END%%" >> $job_to_send
	1305
[251]	1306	elif [[ $remote_host = necriam ]]
[35]	1307	then
	1308	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1309	echo "#PBS -q $return_queue" >> $job_to_send
	1310	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1311	echo "#PBS -j o" >> $job_to_send
	1312	echo " " >> $job_to_send
	1313	echo "set -x" >> $job_to_send
	1314	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1315	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1316	echo "%%END%%" >> $job_to_send
	1317
[181]	1318	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1319	then
	1320	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1321	echo "#!/bin/bash" >> $job_to_send
	1322	echo "#PBS -N job_protocol_transfer" >> $job_to_send
[325]	1323	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[181]	1324	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[366]	1325	echo "#PBS -l feature=data" >> $job_to_send
[181]	1326	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1327	echo "#PBS -j oe" >> $job_to_send
	1328	echo " " >> $job_to_send
	1329	echo ". /usr/share/modules/init/bash" >> $job_to_send
	1330	echo "set -x" >> $job_to_send
	1331	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1332	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1333	echo "%%END%%" >> $job_to_send
	1334
[1]	1335	else
	1336
	1337	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1338	echo "# @\\\$-q $return_queue" >> $job_to_send
	1339	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1340	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1341	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1342	then
	1343	echo "# @\$-l mpp_p=0" >> $job_to_send
	1344	fi
	1345	echo '# @\$-lF 10mb' >> $job_to_send
	1346	# echo '# @\$-o /dev/null' >> $job_to_send
	1347	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1348	echo '# @\\\$-eo' >> $job_to_send
	1349	echo " " >> $job_to_send
	1350	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1351	then
	1352	echo "set +vx" >> $job_to_send
	1353	echo ". .profile" >> $job_to_send
	1354	fi
	1355	echo "set -x" >> $job_to_send
	1356	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	1357	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1358	echo "%%END%%" >> $job_to_send
	1359	fi
	1360
	1361	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1362	then
	1363	echo "llsubmit scpjob.$kennung" >> $job_to_send
[181]	1364	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1365	then
[748]	1366	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1367	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1368	echo "msub scpjob.$kennung" >> $job_to_send
[1]	1369	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1370	then
	1371	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	1372	elif [[ $remote_host = t3es ]]
	1373	then
	1374	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
[1043]	1375	elif [[ $remote_host = lckiaps ]]
	1376	then
	1377	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
	1378	echo "ssh ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
	1379	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
[1]	1380	else
[1040]	1381	echo "$submcom scpjob.$kennung" >> $job_to_send
[1]	1382	fi
[1043]	1383	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
[622]	1384	then
	1385	echo "rm scpjob.$kennung" >> $job_to_send
	1386	fi
[1]	1387	if [[ $remote_host = nech ]]
	1388	then
	1389	echo "cd -" >> $job_to_send
	1390	fi
	1391	else
	1392	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1393	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1394	fi
	1395	echo "set -x" >> $job_to_send
	1396	echo " ' exit" >> $job_to_send
	1397	echo "set -x" >> $job_to_send
	1398	fi
	1399
	1400
	1401
	1402	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	1403	cat $file_to_send >> $job_to_send
[69]	1404
[1]	1405	if [[ $remote_host = ibm ]]
	1406	then
	1407	echo " " >> $job_to_send
	1408	echo "exit" >> $job_to_send
	1409	fi
[635]	1410
	1411	# remove job file
[622]	1412	if [[ $remote_host = lctit \|\| $remote_host = ibmku ]]
[69]	1413	then
	1414	echo " " >> $job_to_send
	1415	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1416	fi
[1]	1417
	1418
	1419
	1420	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	1421	if [[ -z $remote_user ]]
	1422	then
	1423	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	1424	then
	1425	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	1426	fi
	1427	fi
	1428
	1429
	1430
	1431	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	1432	if [[ $no_submit = false ]]
	1433	then
	1434	if [[ $remote_host != $local_host ]]
	1435	then
	1436	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	1437	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	1438	then
[82]	1439	job_catalog_save=$job_catalog
	1440	job_catalog=job_queue
	1441	elif [[ $remote_host = nech ]]
	1442	then
	1443	job_catalog_save=$job_catalog
	1444	job_catalog=/hpf/b/${remote_user}/job_queue
	1445	fi
	1446	if [[ $local_host = decalpha ]]
	1447	then
	1448	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	1449	# PASSWORT AB
	1450	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	1451	elif [[ $remote_host = nech ]]
	1452	then
	1453	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	1454	# TRANSFERIERT WERDEN
	1455	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1456	else
[82]	1457	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1458	fi
[82]	1459	if [[ $? = 1 ]]
	1460	then
	1461	locat=scp; exit
	1462	fi
	1463	if [[ $remote_host = ibms ]]
	1464	then
	1465	job_catalog=$job_catalog_save
	1466	fi
[1]	1467	[[ $verify = true ]] && printf "\n >>> finished\n"
	1468	else
	1469	eval job_catalog=$job_catalog
	1470	cp $job_to_send ${job_catalog}/$job_on_remhost
	1471	fi
	1472
	1473
	1474
	1475	# NQS- BZW. LOADLEVELER-JOB STARTEN
	1476	if [[ $remote_host != $local_host ]]
	1477	then
	1478	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1479
	1480	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
[1]	1481	then
[635]	1482	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
	1483	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
	1484	elif [[ $remote_host = ibmku ]]
	1485	then
	1486	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1]	1487	else
[635]	1488	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1489	fi
[1]	1490
	1491	[[ $verify = true ]] && printf " >>> o.k.\n"
	1492	else
	1493	cd $job_catalog
[210]	1494	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
[1]	1495	then
	1496	eval $submcom $job_on_remhost
[1021]	1497	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb ]]
[108]	1498	then
[635]	1499	chmod u+x $job_on_remhost
	1500	echo "$submcom $job_on_remhost"
[108]	1501	eval $submcom $job_on_remhost
[1]	1502	elif [[ $local_host = nech ]]
	1503	then
	1504	if [[ $queue = default ]]
	1505	then
[799]	1506	eval $submcom $job_on_remhost
[1]	1507	else
[799]	1508	eval $submcom -q $queue $job_on_remhost
[1]	1509	fi
	1510	else
	1511	qsub $job_on_remhost
	1512	fi
[622]	1513
	1514	# Jobfile must not be deleted on lctit/ibmku!! This will be done
	1515	# only at the end of the job.
	1516	if [[ $local_host != lctit && $local_host != ibmku ]]
	1517	then
	1518	rm $job_on_remhost
	1519	fi
[1]	1520	cd - > /dev/null
	1521	fi
	1522	fi
	1523
	1524
	1525
	1526	# ABSCHLUSSARBEITEN
	1527
	1528	if [[ $no_submit = false ]]
	1529	then
	1530	rm $job_to_send
	1531	fi
	1532	[[ $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 |