Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 993

Last change on this file since 993 was 892, checked in by maronga, 13 years ago
bugfixes in parameter file check, adapted for lcxe6 (was lcxt4), new mrun.config parameter %login_init_cmd introduced
Property svn:keywords set to `Id Rev`
File size: 50.3 KB

Rev	Line
[1]	1	#!/bin/ksh
[82]	2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
[169]	3	# $Id: subjob 892 2012-05-02 13:51:44Z hoffmann $
[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
[1]	142
[437]	143
[1]	144	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
[352]	145	delete_dayfile=false
[799]	146	email_notification=none
[122]	147	group_number=none
[1]	148	locat=normal
	149	no_default_queue=none
	150	no_submit=false
	151	job_catalog="~/job_queue"
	152	job_name=none
	153	local_user=$LOGNAME
	154	node_usage=shared
[475]	155	numprocs=0
[1]	156	punkte="..........................................................."
	157	submcom=qsub
	158	queue=default
	159	remote_host=none
	160	remote_user=""
	161	verify=true
	162
	163	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
[696]	164	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
[1]	165	typeset -L20 spalte1
	166	typeset -R40 spalte2
	167	typeset -L60 spalte3
	168
	169
	170
	171	# FEHLERBEHANDLUNG
	172	# BEI EXIT:
	173	trap 'if [[ $locat != normal ]]
	174	then
	175	case $locat in
	176	(option) printf "\n --> available optios can be displayed"
	177	printf " by typing:"
	178	printf "\n \"subjob ?\" \n";;
	179	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
	180	(*) printf "\n +++ unknown error"
	181	printf "\n please inform S. Raasch!\n"
	182	esac
	183	[[ -f $job_to_send ]] && rm $job_to_send
	184	printf "\n\n+++ SUBJOB killed \n\n"
	185	fi' exit
	186
	187
	188	# BEI TERMINAL-BREAK:
	189	trap '[[ -f $job_to_send ]] && rm $job_to_send
	190	printf "\n\n+++ SUBJOB killed \n\n"
	191	exit
	192	' 2
	193
	194
	195
	196
	197	# LOKALEN HOSTNAMEN FESTSTELLEN
	198	local_host=$(hostname)
	199
	200
	201
	202	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
	203	# UEBERHAUPT ZULAESSIG IST
[717]	204	# Note: One of the entries for "lck" or "lckordi" always has to be
	205	# comment out, because the hostname (node*) is same for both machines
[1]	206	case $local_host in
	207	(atmos) local_addres=172.20.25.35; local_host=lcide;;
[102]	208	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
[5]	209	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
[400]	210	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
	211	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
	212	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
	213	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
	214	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
[291]	215	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
[749]	216	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
[12]	217	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
[372]	218	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
[693]	219	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
[108]	220	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
[1]	221	(cs*) local_addres=136.172.44.131; local_host=nech;;
[5]	222	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
[108]	223	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
[367]	224	(node*) local_addres=165.132.26.61 local_host=lck;;
[717]	225	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
[693]	226	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
[101]	227	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
[1]	228	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
	229	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
[5]	230	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
	231	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
[892]	232	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
[400]	233	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
	234	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
	235	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
	236	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
[739]	237	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
[292]	238	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
[210]	239	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
[440]	240	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
[749]	241	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
[5]	242	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
[407]	243	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
[5]	244	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
	245	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
[437]	246	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
[892]	247	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
[1]	248	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
[5]	249	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
	250	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
[205]	251	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
[789]	252	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
[5]	253	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
[622]	254	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
[1]	255	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
[799]	256	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
[1]	257	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
[259]	258	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
[181]	259	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
[196]	260	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
[635]	261	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
[1]	262	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
[622]	263	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
[1]	264	(*) printf "\n +++ \"$local_host\" unknown";
	265	printf "\n please inform S. Raasch!";
	266	locat=parameter; exit;;
	267	esac
	268
	269
	270
	271	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
	272	remote_host=$local_host
	273
	274
	275
	276
	277	# PROZEDUROPTIONEN EINLESEN
[352]	278	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
[1]	279	do
	280	case $option in
	281	(c) job_catalog=$OPTARG;;
	282	(d) delete_dayfile=true;;
	283	(D) no_submit=true;;
[352]	284	(e) email_notification=$OPTARG;;
[125]	285	(g) group_number=$OPTARG;;
[1]	286	(h) remote_host=$OPTARG;;
	287	(m) memory=$OPTARG;;
	288	(n) job_name=$OPTARG;;
	289	(N) node_usage=$OPTARG;;
	290	(O) threads_per_task=$OPTARG;;
	291	(q) no_default_queue=$OPTARG;;
	292	(t) cputime=$OPTARG;;
	293	(T) tasks_per_node=$OPTARG;;
	294	(u) remote_user=$OPTARG;;
	295	(v) verify=false;;
	296	(X) numprocs=$OPTARG;;
	297	(\?) printf "\n +++ Option $OPTARG unknown \n";
	298	locat=option; exit;;
	299	esac
	300	done
	301
	302
	303	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
	304	shift OPTIND-1; file_to_send=$1
	305
	306
	307	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
	308	if [ "$1" = "?" ]
	309	then
	310	(printf "\n *** subjob can be called as follows:\n"
	311	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
	312	printf "\n Description of available options:\n"
	313	printf "\n Option Description Default-Value"
	314	printf "\n -c job-input- and output-catalog ~/job_queue"
	315	printf "\n -d no job-protocol will be created ---"
	316	printf "\n -D only the job-file will be created ---"
	317	printf "\n -h execution host, available hosts: $remote_host"
[693]	318	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
[251]	319	printf "\n lctit, nech, necriam, unics"
[1]	320	printf "\n -m memory demand per process in MByte ---"
	321	printf "\n -n jobname <jobdatei>"
	322	printf "\n -O threads per task (for OpenMP usage) 1"
	323	printf "\n -q job-queue to be used default"
	324	printf "\n -t allowed cpu-time in seconds ---"
	325	printf "\n -T tasks per node (on parallel hosts) ---"
	326	printf "\n -u username on execution host from .netrc"
	327	printf "\n -v no prompt for confirmation ---"
	328	printf "\n -X # of processors (on parallel hosts) 1"
	329	printf "\n "
	330	printf "\n The only possible positional parameter is <jobfile>:"
	331	printf "\n The complete NQS-job must be provided here."
	332	printf "\n <jobfile>=? creates this outline\n\n") \| more
	333	exit
	334	fi
	335
	336
	337
	338	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
	339	if [[ "$file_to_send" = "" ]]
	340	then
	341	printf "\n +++ job-file missing"
	342	locat=parameter; exit
	343	else
	344	if [[ -f $file_to_send ]]
	345	then
	346	true
	347	else
	348	printf "\n +++ job-file: "
	349	printf "\n $file_to_send"
	350	printf "\n does not exist"
	351	locat=parameter; exit
	352	fi
	353	fi
	354
	355
	356
	357	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
	358	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
	359	if [[ $job_name = none ]]
	360	then
	361	job_name=$file_to_send
	362	fi
	363	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
	364	then
	365	printf "\n +++ job-file name: "
	366	printf "\n $job_name"
	367	printf "\n must not contain \"/\"-characters"
	368	locat=parameter; exit
	369	fi
	370
	371
	372
	373
	374	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
	375	# ODER WENN HOST NICHT ANGEGEBEN WURDE
	376	if [[ $remote_host = none ]]
	377	then
	378	printf "\n +++ host missing"
	379	locat=option; exit
	380	else
	381	case $remote_host in
[251]	382	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[312]	383	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[693]	384	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
[622]	385	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
[251]	386	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	387	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
	388	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
[440]	389	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
[540]	390	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
	391	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
[367]	392	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
[717]	393	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
[635]	394	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
[892]	395	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
[437]	396	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
[693]	397	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
[251]	398	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
	399	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
	400	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
	401	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
	402	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
	403	locat=parameter; exit;;
[1]	404	esac
	405	fi
	406
	407
	408	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
	409	if [[ $no_default_queue != none ]]
	410	then
	411	error=false
	412	ndq=$no_default_queue
	413	case $remote_host in
	414	(ibm) case $ndq in
	415	(p690_express\|p690_standard\|p690_long) error=false;;
	416	(*) error=true;;
	417	esac;;
	418	(ibmh) case $ndq in
[312]	419	(no_class) error=false;;
[1]	420	(*) error=true;;
	421	esac;;
[693]	422	(ibmkisti) case $ndq in
	423	(class.32plus\|class.1-2\|class.2-32) error=false;;
	424	(*) error=true;;
	425	esac;;
[622]	426	(ibmku) case $ndq in
	427	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
	428	(*) error=true;;
	429	esac;;
[1]	430	(ibms) case $ndq in
	431	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
	432	(*) error=true;;
	433	esac;;
	434	(ibmy) case $ndq in
	435	(parallel) error=false;;
	436	(*) error=true;;
	437	esac;;
[440]	438	(lckyoto) case $ndq in
	439	(eh\|ph) error=false;;
	440	(*) error=true;;
	441	esac;;
[191]	442	(lcsgib\|lcsgih) case $ndq in
[416]	443	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
[191]	444	(*) error=true;;
	445	esac;;
[1]	446	(lctit) case $ndq in
[635]	447	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
[1]	448	(*) error=true;;
	449	esac;;
	450	(t3eb) case $ndq in
	451	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
	452	(*) error=true;;
	453	esac;;
[251]	454	(necriam) case $ndq in
[253]	455	(SP\|SS\|P6) error=false;;
[251]	456	(*) error=true;;
	457	esac;;
[1]	458	(t3eh) case $ndq in
	459	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
	460	(*) error=true;;
	461	esac;;
	462	(t3ej2\|t3ej5) case $ndq in
	463	(low\|normal\|high) error=false;;
	464	(*) error=true;;
	465	esac;;
	466	(t3es) case $ndq in
	467	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
	468	(*) error=true;;
	469	esac;;
	470	(unics) case $ndq in
	471	(unics\|ht) error=false;;
	472	(*) error=true;;
	473	esac;;
	474	esac
	475	if [[ $error = true ]]
	476	then
	477	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
	478	locat=parameter; exit
	479	else
	480	queue=$no_default_queue
	481	fi
	482	fi
	483
	484
	485
	486	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
	487	# AUFTEILEN
	488	done=false
	489	while [[ $done = false ]]
	490	do
	491	if (( $cputime <= 0 ))
	492	then
	493	printf "\n +++ wrong cpu-time or cpu-time missing"
	494	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
	495	printf "\n >>> "
	496	read cputime 1>/dev/null 2>&1
	497	else
	498	done=true
	499	fi
	500	done
	501	if [[ $remote_host = nech ]]
	502	then
	503	if (( tasks_per_node != 0 ))
	504	then
	505	(( cputime = cputime * tasks_per_node ))
	506	elif [[ $numprocs != 0 ]]
	507	then
	508	(( cputime = cputime * numprocs ))
	509	fi
	510	fi
	511	(( stunden = cputime / 3600 ))
	512	(( resttime = cputime - stunden * 3600 ))
	513	(( minuten = resttime / 60 ))
	514	(( sekunden = resttime - minuten * 60 ))
	515	timestring=${stunden}:${minuten}:${sekunden}
	516
	517
	518
	519	# PRUEFEN DER KERNSPEICHERANFORDERUNG
	520	done=false
	521	while [[ $done = false ]]
	522	do
	523	if (( memory <= 0 ))
	524	then
	525	printf "\n +++ wrong memory demand or memory demand missing"
	526	printf "\n >>> Please type memory in MByte per process as INTEGER:"
	527	printf "\n >>> "
	528	read memory 1>/dev/null 2>&1
	529	else
	530	done=true
	531	fi
	532	done
	533
[251]	534	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
[1]	535	then
	536	if (( tasks_per_node != 0 ))
	537	then
	538	(( Memory = memory * tasks_per_node / 1000 ))
	539	elif [[ $numprocs != 0 ]]
	540	then
	541	(( Memory = memory * numprocs / 1000 ))
	542	else
	543	(( Memory = memory / 1000 ))
	544	fi
[635]	545	elif [[ $remote_host = lctit ]]
	546	then
	547	(( Memory = memory * tasks_per_node / 1000 ))
[1]	548	fi
	549
	550
	551	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
	552	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	553	then
	554	(( memory = memory * threads_per_task ))
	555	fi
	556
	557
	558	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
	559	if (( tasks_per_node != 0 ))
	560	then
[622]	561	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
[1]	562	fi
	563
[622]	564	# Calculate number of processes per node
	565	(( processes_per_node = tasks_per_node * threads_per_task ))
[1]	566
[696]	567	# Calculate number of MPI tasks
	568	(( mpi_tasks = numprocs / threads_per_task ))
[1]	569
[696]	570
[1]	571	# HEADER-AUSGABE
	572	if [[ $verify = true ]]
	573	then
	574	printf "\n\n"
	575	printf "#--------------------------------------------------------------# \n"
	576	spalte1=SUBJOB;spalte2=$(date)
	577	printf "\| $spalte1$spalte2 \| \n"
	578	printf "\| \| \n"
	579	printf "\| values of parameters/options: \| \n"
	580	spalte1=local_host$punkte; spalte2=$punkte$local_host
	581	printf "\| $spalte1$spalte2 \| \n"
	582	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
	583	printf "\| $spalte1$spalte2 \| \n"
	584	spalte1=queue$punkte; spalte2=$punkte$queue
	585	printf "\| $spalte1$spalte2 \| \n"
	586	spalte1=memory$punkte; spalte2="$punkte$memory mb"
	587	printf "\| $spalte1$spalte2 \| \n"
	588	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
	589	printf "\| $spalte1$spalte2 \| \n"
	590	spalte1=job_name$punkte; spalte2="$punkte$job_name"
	591	printf "\| $spalte1$spalte2 \| \n"
	592	printf "#--------------------------------------------------------------# \n\n"
	593
	594
	595	# KONTROLLABFRAGE, OB ALLES O.K.
	596	antwort="dummy"
	597	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
	598	do
	599	read antwort?" >>> continue (y/n) ? "
	600	done
	601	if [[ $antwort = n \|\| $antwort = N ]]
	602	then
	603	locat=verify; exit
	604	fi
	605	printf "\n"
	606	fi
	607
	608
	609
	610	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
	611	kennung=$RANDOM
	612	job_on_remhost=${job_name}_${kennung}_$local_host
	613	job_to_send=job_to_send_$kennung
	614	if [[ $delete_dayfile = false ]]
	615	then
	616	remote_dayfile=${local_host}_${job_name}_result_$kennung
	617	local_dayfile=${remote_host}_${job_name}
	618	else
	619	remote_dayfile=/dev/null
	620	fi
	621
	622
	623
[622]	624	# Generate the batch job scripts (qsub/msub/LoadLeveler)
[1]	625	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
	626	then
	627
[622]	628	# General LoadLeveler settings
	629	execute_in_shell="#!/bin/ksh"
	630	use_shell="# @ shell = /bin/ksh"
	631	consumable_memory="ConsumableMemory($memory mb)"
	632	class="# @ class = $queue"
	633	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
	634	network_to_use="# @ network.mpi = sn_all,shared,us"
	635	data_limit="# @ data_limit = 1.76gb"
	636	image_size="# @ image_size = 50"
[693]	637	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
[312]	638
[693]	639	if [[ $email_notification = none ]]
	640	then
	641	notify_user=""
	642	else
	643	notify_user="# @ notify_user = $email_notification"
	644	if [[ $delete_dayfile = true ]]
	645	then
	646	notification='# @ notification = never'
	647	fi
	648	fi
[622]	649
[312]	650	if [[ $remote_host = ibmh ]]
[1]	651	then
[312]	652	data_limit=""
	653	network_to_use=""
	654	class=""
	655	environment=""
[814]	656	rset="# @ rset = RSET_MCM_AFFINITY"
	657	task_affinity="# @ task_affinity = core(1)"
[693]	658	elif [[ $remote_host = ibmkisti ]]
	659	then
	660	network_to_use="# @ network.MPI = sn_all,shared,US"
	661	wall_clock_limit="# @ wall_clock_limit = $timestring"
[696]	662	if [[ $threads_per_task = 1 ]]
	663	then
	664	rset="# @ rset = RSET_MCM_AFFINITY"
	665	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
	666	fi
[693]	667	environment=""
	668	use_shell=""
	669	data_limit=""
	670	image_size=""
[622]	671	elif [[ $remote_host = ibmku ]]
	672	then
	673	execute_in_shell="#!/usr/bin/ksh"
	674	use_shell="# @ shell = /usr/bin/ksh"
	675	consumable_memory=""
	676	environment=""
	677	network_to_use="# @ network.mpi = sn_all,shared,us"
	678	data_limit=""
	679	image_size=""
	680	elif [[ $remote_host = ibms ]]
	681	then
	682	network_to_use="# @ network.mpi = csss,shared,us"
	683	elif [[ $remote_host = ibmy ]]
	684	then
	685	consumable_memory=""
	686	network_to_use=""
[1]	687	fi
	688
	689	cat > $job_to_send << %%END%%
[622]	690	$execute_in_shell
	691	$use_shell
[1]	692
	693	# @ job_type = parallel
[693]	694	# @ job_name = $job_name
[1]	695	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
	696	# @ output = $remote_dayfile
	697	# @ error = $remote_dayfile
[693]	698	$wall_clock_limit
[622]	699	$image_size
[312]	700	$class
	701	$environment
[1]	702	$network_to_use
	703	$data_limit
[693]	704	$rset
	705	$mcm_affinity_options
[814]	706	$task_affinity
[1]	707	$notification
[693]	708	$notify_user
[1]	709
	710	%%END%%
	711
	712	if (( nodes > 0 ))
	713	then
	714
[693]	715	if [[ $remote_host != ibmkisti ]]
	716	then
	717
	718	cat >> $job_to_send << %%END%%
[1]	719	# @ node = $nodes
[622]	720	# @ tasks_per_node = $processes_per_node
[1]	721	# @ node_usage = $node_usage
	722	# @ queue
	723
	724	%%END%%
	725
[693]	726	else
	727
	728	cat >> $job_to_send << %%END%%
[696]	729	# @ total_tasks = $mpi_tasks
[693]	730	# @ blocking = unlimited
	731	# @ queue
	732
	733	%%END%%
	734
	735	fi
	736
[1]	737	else
	738
	739	if [[ $remote_host != ibmy ]]
	740	then
	741
	742	cat >> $job_to_send << %%END%%
	743	# @ blocking = unlimited
	744	# @ total_tasks = $numprocs
	745	# @ node_usage = $node_usage
	746	# @ queue
	747
	748	%%END%%
	749
	750	else
	751
	752	cat >> $job_to_send << %%END%%
	753	# @ node = 1
	754	# @ total_tasks = $numprocs
	755	# @ queue
	756
	757	%%END%%
	758
	759	fi
	760
	761	fi
	762
[696]	763	# workaround because of silly job filter on ibmkisti
	764	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
	765	then
	766	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
	767	fi
	768
[1]	769	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
	770	then
	771
	772	cat > $job_to_send << %%END%%
	773	#!/bin/ksh
	774
	775	# @ job_type = serial
	776	# @ node_usage = $node_usage
	777	# @ job_name = palm
	778	# @ wall_clock_limit = ${timestring},$timestring
	779	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
	780	# @ output = $remote_dayfile
	781	# @ error = $remote_dayfile
[312]	782	$class
[1]	783	$notification
	784
	785	# @ queue
	786
	787	%%END%%
	788
[108]	789	elif [[ $remote_host = lcfimm ]]
	790	then
	791
	792	if [[ $numprocs != 0 ]]
	793	then
	794	cat > $job_to_send << %%END%%
	795	#!/bin/ksh
	796	#PBS -N $job_name
[552]	797	#PBS -A $project_account
[108]	798	#PBS -l walltime=$timestring
[622]	799	#PBS -l nodes=${nodes}:ppn=$processes_per_node
[108]	800	#PBS -l pmem=${memory}mb
	801	#PBS -m abe
	802	#PBS -o $remote_dayfile
	803	#PBS -j oe
	804	mpd &
	805
	806	%%END%%
	807
[168]	808	else
	809	cat > $job_to_send << %%END%%
	810	#!/bin/ksh
	811	#PBS -N $job_name
[552]	812	#PBS -A $project_account
[168]	813	#PBS -l walltime=$timestring
	814	#PBS -l ncpus=1
	815	#PBS -l pmem=${memory}mb
	816	#PBS -m abe
	817	#PBS -o $remote_dayfile
	818	#PBS -j oe
	819
	820	%%END%%
	821
	822	fi
	823
[717]	824	elif [[ $remote_host = lck \|\| $remote_host = lckordi ]]
[368]	825	then
	826
	827	if [[ $numprocs != 0 ]]
	828	then
	829	cat > $job_to_send << %%END%%
	830	#!/bin/ksh
	831	#PBS -N $job_name
	832	#PBS -l walltime=$timestring
	833	#PBS -l ncpus=$numprocs
	834	#PBS -l pmem=${memory}mb
	835	#PBS -o $remote_dayfile
	836	#PBS -j oe
	837
	838	mpd &
	839
	840	%%END%%
	841
	842	else
	843	cat > $job_to_send << %%END%%
	844	#!/bin/ksh
	845	#PBS -N $job_name
	846	#PBS -l walltime=$timestring
	847	#PBS -l ncpus=1
	848	#PBS -l pmem=${memory}mb
	849	#PBS -o $remote_dayfile
	850	#PBS -j oe
	851
	852	%%END%%
	853
	854	fi
	855
[693]	856	elif [[ $remote_host = lcyon ]]
	857	then
	858
	859	if [[ $numprocs != 0 ]]
	860	then
	861	cat > $job_to_send << %%END%%
	862	#!/bin/ksh
	863	#PBS -N $job_name
	864	#PBS -l walltime=$timestring
	865	#PBS -l ncpus=$numprocs
	866	#PBS -l pmem=${memory}mb
	867	#PBS -o $remote_dayfile
	868	#PBS -j oe
	869
	870	%%END%%
	871
	872	else
	873	cat > $job_to_send << %%END%%
	874	#!/bin/ksh
	875	#PBS -N $job_name
	876	#PBS -l walltime=$timestring
	877	#PBS -l ncpus=1
	878	#PBS -l pmem=${memory}mb
	879	#PBS -o $remote_dayfile
	880	#PBS -j oe
	881
	882	%%END%%
	883
	884	fi
	885
[181]	886	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
[179]	887	then
	888
[376]	889	if [[ "$sgi_feature" != "" ]]
	890	then
	891	feature_directive="#PBS -l feature=$sgi_feature"
	892	else
	893	feature_directive=""
	894	fi
[799]	895
[416]	896	if [[ $queue = dataq \|\| $queue = permq ]]
[399]	897	then
	898	feature_directive="#PBS -l feature=data"
	899	fi
[376]	900
[789]	901	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
[191]	902	then
	903	queue_directive="#PBS -q $queue"
	904	else
	905	queue_directive=""
	906	fi
	907
[352]	908	if [[ $email_notification = none ]]
	909	then
	910	email_directive=""
	911	else
	912	email_directive="#PBS -M $email_notification"
	913	fi
	914
[179]	915	if [[ $numprocs != 0 ]]
	916	then
	917	cat > $job_to_send << %%END%%
[181]	918	#!/bin/bash
[179]	919	#PBS -N $job_name
	920	#PBS -l walltime=$timestring
[622]	921	#PBS -l nodes=$nodes:ppn=${processes_per_node}
[181]	922	#PBS -l naccesspolicy=$node_usage
[179]	923	#PBS -o $remote_dayfile
	924	#PBS -j oe
[376]	925	$feature_directive
[191]	926	$queue_directive
[352]	927	$email_directive
[179]	928
[622]	929	eval \`/sw/swdist/bin/modulesinit\`
	930	#. /usr/share/modules/init/bash
[892]	931	$init_cmds
[493]	932	$module_calls
[374]	933
[181]	934	echo ld_library_path=\$LD_LIBRARY_PATH
[179]	935
	936	%%END%%
	937
	938	else
	939	cat > $job_to_send << %%END%%
	940	#PBS -S /bin/bash
	941	#PBS -N $job_name
	942	#PBS -l walltime=$timestring
	943	#PBS -l ncpus=1
	944	#PBS -o $remote_dayfile
[475]	945	#PBS -j oe
[400]	946	$queue_directive
	947	$email_directive
[179]	948
[622]	949	eval \`/sw/swdist/bin/modulesinit\`
	950	#. /usr/share/modules/init/bash
[892]	951	$init_cmds
[493]	952	$module_calls
[475]	953
[179]	954	%%END%%
	955
	956	fi
	957
[892]	958	elif [[ $remote_host = lcxe6 ]]
[164]	959	then
	960
	961	if [[ $numprocs != 0 ]]
	962	then
	963	cat > $job_to_send << %%END%%
[799]	964	#!/bin/ksh
[164]	965	#PBS -S /bin/ksh
	966	#PBS -N $job_name
[552]	967	#PBS -A $project_account
[206]	968	#PBS -j oe
[164]	969	#PBS -l walltime=$timestring
	970	#PBS -l mppwidth=${numprocs}
[622]	971	#PBS -l mppnppn=${processes_per_node}
[164]	972	#PBS -m abe
	973	#PBS -o $remote_dayfile
[492]	974	$email_directive
[164]	975
[892]	976	$init_cmds
[493]	977	$module_calls
[343]	978
[164]	979	%%END%%
	980
[108]	981	else
	982	cat > $job_to_send << %%END%%
	983	#!/bin/ksh
[168]	984	#PBS -S /bin/ksh
[108]	985	#PBS -N $job_name
[552]	986	#PBS -A $project_account
[206]	987	#PBS -j oe
[108]	988	#PBS -l walltime=$timestring
	989	#PBS -l ncpus=1
	990	#PBS -l pmem=${memory}mb
	991	#PBS -m abe
[492]	992	$email_directive
[108]	993	#PBS -o $remote_dayfile
	994
[892]	995	$init_cmds
[493]	996	$module_calls
[343]	997
[108]	998	%%END%%
	999
	1000	fi
	1001
[440]	1002	elif [[ $remote_host = lckyoto ]]
[437]	1003	then
	1004
[440]	1005	cat > $job_to_send << %%END%%
[799]	1006	#!/bin/ksh
[440]	1007	# @\$-o $remote_dayfile
	1008	# @\$-eo -oi
	1009	# @\$-lP 16
[799]	1010	# @\$-lp 1
[440]	1011	# @\$-lm 28gb -llm unlimited -ls unlimited
	1012	# @\$-q $queue
	1013	# @\$-Pvn abs_pack
	1014	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
	1015	#. /thin/local/etc/setprofile/intel-11.0.sh
	1016	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
	1017	. ~/.myprofile
	1018	#. /home2/t/t51254/palm/current_version/myprofile
	1019	#. /thin/apps/pgi/mpi.sh
	1020	#
	1021	env
	1022	#
	1023	set -x
	1024
	1025	%%END%%
	1026
	1027	elif [[ $remote_host = lcxt5m ]]
	1028	then
	1029
[437]	1030	if [[ $numprocs != 0 ]]
	1031	then
	1032	cat > $job_to_send << %%END%%
[799]	1033	#!/bin/ksh
[437]	1034	#PBS -S /bin/ksh
	1035	#PBS -N $job_name
	1036	#PBS -j oe
	1037	#PBS -l walltime=$timestring
	1038	#PBS -l mppwidth=${numprocs}
[622]	1039	#PBS -l mppnppn=${processes_per_node}
[437]	1040	#PBS -m abe
	1041	#PBS -o $remote_dayfile
	1042
[892]	1043	$init_cmds
[493]	1044	$module_calls
[437]	1045
	1046	%%END%%
	1047
	1048	else
	1049	cat > $job_to_send << %%END%%
	1050	#!/bin/ksh
	1051	#PBS -S /bin/ksh
	1052	#PBS -N $job_name
	1053	#PBS -j oe
	1054	#PBS -l walltime=$timestring
	1055	#PBS -l ncpus=1
	1056	#PBS -l pmem=${memory}mb
	1057	#PBS -m abe
	1058	#PBS -o $remote_dayfile
	1059
[892]	1060	$init_cmds
[493]	1061	$module_calls
[437]	1062
	1063	%%END%%
	1064
	1065	fi
	1066
[1]	1067	elif [[ $remote_host = nech ]]
	1068	then
	1069
	1070	if (( nodes > 1 ))
	1071	then
	1072	cat > $job_to_send << %%END%%
	1073	#!/bin/ksh
[622]	1074	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1075	#PBS -l ${qsubmem}=${Memory}gb
	1076	#PBS -b $nodes
	1077	#PBS -o $remote_dayfile
	1078	#PBS -N palm
	1079	#PBS -j o
	1080	#PBS -T mpisx
	1081
	1082	%%END%%
	1083
	1084	elif [[ $numprocs != 0 ]]
	1085	then
	1086	cat > $job_to_send << %%END%%
	1087	#!/bin/ksh
[622]	1088	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
[1]	1089	#PBS -l ${qsubmem}=${Memory}gb
	1090	#PBS -o $remote_dayfile
	1091	#PBS -N palm
	1092	#PBS -j o
	1093
	1094	%%END%%
	1095
	1096	else
	1097	cat > $job_to_send << %%END%%
	1098	#!/bin/ksh
	1099	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
	1100	#PBS -o $remote_dayfile
	1101	#PBS -j o
	1102
	1103	%%END%%
	1104
	1105	fi
	1106
[251]	1107	elif [[ $remote_host = necriam ]]
[35]	1108	then
	1109
	1110	if (( nodes > 1 ))
	1111	then
	1112	cat > $job_to_send << %%END%%
	1113	#!/bin/ksh
	1114	#PBS -b $nodes
	1115	#PBS -o $remote_dayfile
[253]	1116	#PBS -N $job_name
[35]	1117	#PBS -j o
[251]	1118	#PBS -v MPIPROGINV=YES
[35]	1119
	1120	%%END%%
	1121
	1122	elif [[ $numprocs != 0 ]]
	1123	then
	1124	cat > $job_to_send << %%END%%
	1125	#!/bin/ksh
	1126	#PBS -q ${queue}
	1127	#PBS -o $remote_dayfile
[253]	1128	#PBS -N $job_name
[35]	1129	#PBS -j o
[251]	1130	#PBS -v MPIPROGINV=YES
[35]	1131
	1132	%%END%%
	1133
	1134	else
	1135	cat > $job_to_send << %%END%%
	1136	#!/bin/ksh
	1137	#PBS -o $remote_dayfile
	1138	#PBS -j o
	1139
	1140	%%END%%
	1141
	1142	fi
	1143
[1]	1144	elif [[ $remote_host = lctit ]]
	1145	then
	1146	cat > $job_to_send << %%END%%
[635]	1147	#!/bin/ksh
[892]	1148	$init_cmds
[678]	1149	$module_calls
	1150
[1]	1151	%%END%%
	1152
	1153	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
[678]	1154	if [[ $tasks_per_node != $processes_per_node ]]
	1155	then
	1156	submcom="$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 $queue "
	1157	else
	1158	submcom="$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 "
	1159	fi
[1]	1160
	1161	else
	1162
	1163	cat > $job_to_send << %%END%%
	1164	# @\$-q ${queue}
	1165	# @\$-l${qsubtime} $timestring
	1166	# @\$-l${qsubmem} ${memory}mb
	1167	# @\$-o $remote_dayfile
	1168	# @\$-eo
	1169
	1170	%%END%%
	1171
	1172	fi
	1173
	1174
	1175	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
	1176	# BEI EXIT VERANLASSEN
	1177	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
	1178	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
	1179	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
	1180	then
	1181	echo "set +vx" >> $job_to_send
	1182	echo "trap '" >> $job_to_send
	1183	echo "set +vx" >> $job_to_send
[635]	1184	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec ]]
[1]	1185	then
[622]	1186	if [[ $remote_host = ibmh ]]
[1]	1187	then
	1188	return_queue=c1
[693]	1189	elif [[ $remote_host = ibmkisti ]]
	1190	then
	1191	return_queue=class.1-2
[622]	1192	elif [[ $remote_host = ibmku ]]
	1193	then
	1194	return_queue=sdbg2
[1]	1195	elif [[ $remote_host = ibms ]]
	1196	then
	1197	return_queue=p_normal
	1198	elif [[ $remote_host = ibmy ]]
	1199	then
	1200	return_queue=serial
[325]	1201	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
	1202	then
[326]	1203	return_queue=serialq
[251]	1204	elif [[ $remote_host = necriam ]]
[1]	1205	then
[253]	1206	return_queue=SP
[1]	1207	else
	1208	return_queue=unknown
	1209	fi
	1210
	1211	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1212	then
	1213
[622]	1214	if [[ $remote_host = ibmku ]]
	1215	then
	1216	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1217	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1218	else
	1219	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
	1220	fi
[1]	1221	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
	1222	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
	1223	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
	1224	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
	1225	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
	1226	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
[312]	1227	if [[ $host != "ibmh" ]]
	1228	then
	1229	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
	1230	fi
[1]	1231	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
	1232	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
	1233
	1234	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
	1235	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
	1236
	1237	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
	1238	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]	1239	if [[ $remote_host = ibmku ]]
	1240	then
	1241	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
	1242	fi
[1]	1243	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
	1244
	1245	elif [[ $remote_host = nech ]]
	1246	then
	1247	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1248	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1249	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
	1250	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1251	echo "#PBS -j o" >> $job_to_send
	1252	echo " " >> $job_to_send
	1253	echo "set -x" >> $job_to_send
	1254	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
	1255	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1256	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1257	echo "%%END%%" >> $job_to_send
	1258
[251]	1259	elif [[ $remote_host = necriam ]]
[35]	1260	then
	1261	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1262	echo "#PBS -q $return_queue" >> $job_to_send
	1263	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
	1264	echo "#PBS -j o" >> $job_to_send
	1265	echo " " >> $job_to_send
	1266	echo "set -x" >> $job_to_send
	1267	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$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
[181]	1271	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1272	then
	1273	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1274	echo "#!/bin/bash" >> $job_to_send
	1275	echo "#PBS -N job_protocol_transfer" >> $job_to_send
[325]	1276	echo "#PBS -l walltime=00:30:00" >> $job_to_send
[181]	1277	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
[366]	1278	echo "#PBS -l feature=data" >> $job_to_send
[181]	1279	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1280	echo "#PBS -j oe" >> $job_to_send
	1281	echo " " >> $job_to_send
	1282	echo ". /usr/share/modules/init/bash" >> $job_to_send
	1283	echo "set -x" >> $job_to_send
	1284	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1285	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1286	echo "%%END%%" >> $job_to_send
	1287
[1]	1288	else
	1289
	1290	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
	1291	echo "# @\\\$-q $return_queue" >> $job_to_send
	1292	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
	1293	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
	1294	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
	1295	then
	1296	echo "# @\$-l mpp_p=0" >> $job_to_send
	1297	fi
	1298	echo '# @\$-lF 10mb' >> $job_to_send
	1299	# echo '# @\$-o /dev/null' >> $job_to_send
	1300	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
	1301	echo '# @\\\$-eo' >> $job_to_send
	1302	echo " " >> $job_to_send
	1303	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1304	then
	1305	echo "set +vx" >> $job_to_send
	1306	echo ". .profile" >> $job_to_send
	1307	fi
	1308	echo "set -x" >> $job_to_send
	1309	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
	1310	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
	1311	echo "%%END%%" >> $job_to_send
	1312	fi
	1313
	1314	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
	1315	then
	1316	echo "llsubmit scpjob.$kennung" >> $job_to_send
[181]	1317	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
	1318	then
[748]	1319	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
	1320	echo "chmod u+x scpjob.$kennung" >> $job_to_send
	1321	echo "msub scpjob.$kennung" >> $job_to_send
[1]	1322	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
	1323	then
	1324	echo "qsub -J n scpjob.$kennung" >> $job_to_send
	1325	elif [[ $remote_host = t3es ]]
	1326	then
	1327	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
	1328	else
	1329	echo "qsub scpjob.$kennung" >> $job_to_send
	1330	fi
[622]	1331	if [[ $remote_host != ibmku ]]
	1332	then
	1333	echo "rm scpjob.$kennung" >> $job_to_send
	1334	fi
[1]	1335	if [[ $remote_host = nech ]]
	1336	then
	1337	echo "cd -" >> $job_to_send
	1338	fi
	1339	else
	1340	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
	1341	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
	1342	fi
	1343	echo "set -x" >> $job_to_send
	1344	echo " ' exit" >> $job_to_send
	1345	echo "set -x" >> $job_to_send
	1346	fi
	1347
	1348
	1349
	1350	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
	1351	cat $file_to_send >> $job_to_send
[69]	1352
[1]	1353	if [[ $remote_host = ibm ]]
	1354	then
	1355	echo " " >> $job_to_send
	1356	echo "exit" >> $job_to_send
	1357	fi
[635]	1358
	1359	# remove job file
[622]	1360	if [[ $remote_host = lctit \|\| $remote_host = ibmku ]]
[69]	1361	then
	1362	echo " " >> $job_to_send
	1363	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
	1364	fi
[1]	1365
	1366
	1367
	1368	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
	1369	if [[ -z $remote_user ]]
	1370	then
	1371	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
	1372	then
	1373	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
	1374	fi
	1375	fi
	1376
	1377
	1378
	1379	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
	1380	if [[ $no_submit = false ]]
	1381	then
	1382	if [[ $remote_host != $local_host ]]
	1383	then
	1384	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
[82]	1385	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
[1]	1386	then
[82]	1387	job_catalog_save=$job_catalog
	1388	job_catalog=job_queue
	1389	elif [[ $remote_host = nech ]]
	1390	then
	1391	job_catalog_save=$job_catalog
	1392	job_catalog=/hpf/b/${remote_user}/job_queue
	1393	fi
	1394	if [[ $local_host = decalpha ]]
	1395	then
	1396	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
	1397	# PASSWORT AB
	1398	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
	1399	elif [[ $remote_host = nech ]]
	1400	then
	1401	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
	1402	# TRANSFERIERT WERDEN
	1403	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
[1]	1404	else
[82]	1405	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
[1]	1406	fi
[82]	1407	if [[ $? = 1 ]]
	1408	then
	1409	locat=scp; exit
	1410	fi
	1411	if [[ $remote_host = ibms ]]
	1412	then
	1413	job_catalog=$job_catalog_save
	1414	fi
[1]	1415	[[ $verify = true ]] && printf "\n >>> finished\n"
	1416	else
	1417	eval job_catalog=$job_catalog
	1418	cp $job_to_send ${job_catalog}/$job_on_remhost
	1419	fi
	1420
	1421
	1422
	1423	# NQS- BZW. LOADLEVELER-JOB STARTEN
	1424	if [[ $remote_host != $local_host ]]
	1425	then
	1426	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
[635]	1427
	1428	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
[1]	1429	then
[635]	1430	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
	1431	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
	1432	elif [[ $remote_host = ibmku ]]
	1433	then
	1434	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
[1]	1435	else
[635]	1436	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
[82]	1437	fi
[1]	1438
	1439	[[ $verify = true ]] && printf " >>> o.k.\n"
	1440	else
	1441	cd $job_catalog
[210]	1442	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
[1]	1443	then
	1444	eval $submcom $job_on_remhost
[892]	1445	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi\|\| $localhost = lcyon ]]
[108]	1446	then
[635]	1447	chmod u+x $job_on_remhost
	1448	echo "$submcom $job_on_remhost"
[108]	1449	eval $submcom $job_on_remhost
[1]	1450	elif [[ $local_host = nech ]]
	1451	then
	1452	if [[ $queue = default ]]
	1453	then
[799]	1454	eval $submcom $job_on_remhost
[1]	1455	else
[799]	1456	eval $submcom -q $queue $job_on_remhost
[1]	1457	fi
	1458	else
	1459	qsub $job_on_remhost
	1460	fi
[622]	1461
	1462	# Jobfile must not be deleted on lctit/ibmku!! This will be done
	1463	# only at the end of the job.
	1464	if [[ $local_host != lctit && $local_host != ibmku ]]
	1465	then
	1466	rm $job_on_remhost
	1467	fi
[1]	1468	cd - > /dev/null
	1469	fi
	1470	fi
	1471
	1472
	1473
	1474	# ABSCHLUSSARBEITEN
	1475
	1476	if [[ $no_submit = false ]]
	1477	then
	1478	rm $job_to_send
	1479	fi
	1480	[[ $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 |