Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1094

Last change on this file since 1094 was 1094, checked in by raasch, 12 years ago
scp/ssh port can be set explicitly
Property svn:keywords set to `Id Rev`
File size: 48.1 KB

Line
1	#!/bin/ksh
2
3	# subjob - script for automatic generation and submission of batch-job files
4	# for various batch queuing systems
5
6	#--------------------------------------------------------------------------------#
7	# This file is part of PALM.
8	#
9	# PALM is free software: you can redistribute it and/or modify it under the terms
10	# of the GNU General Public License as published by the Free Software Foundation,
11	# either version 3 of the License, or (at your option) any later version.
12	#
13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
16	#
17	# You should have received a copy of the GNU General Public License along with
18	# PALM. If not, see <http://www.gnu.org/licenses/>.
19	#
20	# Copyright 1997-2012 Leibniz University Hannover
21	#--------------------------------------------------------------------------------#
22	#
23	# Current revisions:
24	# ------------------
25	# new option -P for explicit setting of ssh/scp port
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1094 2013-02-03 01:52:12Z raasch $
30	#
31	# 2013-02-02 07:06:13Z raasch
32	# adjustments for Kyushu-University computing center (lckyut)
33	# old changelog messages removed
34	#
35	# 1046 2012-11-09 14:38:45Z maronga
36	# code put under GPL (PALM 3.9)
37	#
38	# 08/07/94 - Siggi - first version finished
39	# 29/06/94 - Siggi - script development started
40	#--------------------------------------------------------------------------------#
41	# subjob - script for automatic generation and submission of batch-job files
42	# for various batch queuing systems
43	#--------------------------------------------------------------------------------#
44
45
46	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
47	delete_dayfile=false
48	email_notification=none
49	group_number=none
50	locat=normal
51	no_default_queue=none
52	no_submit=false
53	job_catalog="~/job_queue"
54	job_name=none
55	local_user=$LOGNAME
56	node_usage=shared
57	numprocs=0
58	punkte="..........................................................."
59	submcom=qsub
60	queue=default
61	remote_host=none
62	remote_user=""
63	verify=true
64
65	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
66	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
67	typeset -L20 spalte1
68	typeset -R40 spalte2
69	typeset -L60 spalte3
70
71
72
73	# FEHLERBEHANDLUNG
74	# BEI EXIT:
75	trap 'if [[ $locat != normal ]]
76	then
77	case $locat in
78	(option) printf "\n --> available optios can be displayed"
79	printf " by typing:"
80	printf "\n \"subjob ?\" \n";;
81	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
82	(*) printf "\n +++ unknown error"
83	printf "\n please inform S. Raasch!\n"
84	esac
85	[[ -f $job_to_send ]] && rm $job_to_send
86	printf "\n\n+++ SUBJOB killed \n\n"
87	fi' exit
88
89
90	# BEI TERMINAL-BREAK:
91	trap '[[ -f $job_to_send ]] && rm $job_to_send
92	printf "\n\n+++ SUBJOB killed \n\n"
93	exit
94	' 2
95
96
97
98
99	# LOKALEN HOSTNAMEN FESTSTELLEN
100	local_host=$(hostname)
101
102
103
104	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
105	# UEBERHAUPT ZULAESSIG IST
106	# Note: One of the entries for "lck" or "lckordi" always has to be
107	# comment out, because the hostname (node*) is same for both machines
108	case $local_host in
109	(atmos) local_addres=172.20.25.35; local_host=lcide;;
110	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
111	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
112	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
113	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
114	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
115	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
116	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
117	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
118	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
119	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
120	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
121	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
122	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
123	(cs*) local_addres=136.172.44.131; local_host=nech;;
124	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
125	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
126	(node*) local_addres=165.132.26.61 local_host=lck;;
127	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
128	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
129	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
130	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
131	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
132	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
133	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
134	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
135	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
136	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
137	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
138	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
139	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
140	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
141	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
142	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
143	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
144	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
145	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
146	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
147	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
148	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
149	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
150	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
151	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
152	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
153	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
154	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
155	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
156	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
157	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
158	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
159	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
160	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
161	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
162	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
163	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
164	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
165	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
166	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
167	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
168	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
169	(*) printf "\n +++ \"$local_host\" unknown";
170	printf "\n please inform S. Raasch!";
171	locat=parameter; exit;;
172	esac
173
174
175
176	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
177	remote_host=$local_host
178
179
180
181
182	# PROZEDUROPTIONEN EINLESEN
183	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
184	do
185	case $option in
186	(c) job_catalog=$OPTARG;;
187	(d) delete_dayfile=true;;
188	(D) no_submit=true;;
189	(e) email_notification=$OPTARG;;
190	(g) group_number=$OPTARG;;
191	(h) remote_host=$OPTARG;;
192	(m) memory=$OPTARG;;
193	(n) job_name=$OPTARG;;
194	(N) node_usage=$OPTARG;;
195	(O) threads_per_task=$OPTARG;;
196	(P) scp_port=$OPTARG;;
197	(q) no_default_queue=$OPTARG;;
198	(t) cputime=$OPTARG;;
199	(T) tasks_per_node=$OPTARG;;
200	(u) remote_user=$OPTARG;;
201	(v) verify=false;;
202	(X) numprocs=$OPTARG;;
203	(\?) printf "\n +++ Option $OPTARG unknown \n";
204	locat=option; exit;;
205	esac
206	done
207
208
209	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
210	shift OPTIND-1; file_to_send=$1
211
212
213	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
214	if [ "$1" = "?" ]
215	then
216	(printf "\n *** subjob can be called as follows:\n"
217	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
218	printf "\n Description of available options:\n"
219	printf "\n Option Description Default-Value"
220	printf "\n -c job-input- and output-catalog ~/job_queue"
221	printf "\n -d no job-protocol will be created ---"
222	printf "\n -D only the job-file will be created ---"
223	printf "\n -h execution host, available hosts: $remote_host"
224	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
225	printf "\n lckiaps, lctit, nech, necriam, unics"
226	printf "\n -m memory demand per process in MByte ---"
227	printf "\n -n jobname <jobdatei>"
228	printf "\n -O threads per task (for OpenMP usage) 1"
229	printf "\n -P ssh/scp port default port"
230	printf "\n -q job-queue to be used default"
231	printf "\n -t allowed cpu-time in seconds ---"
232	printf "\n -T tasks per node (on parallel hosts) ---"
233	printf "\n -u username on execution host from .netrc"
234	printf "\n -v no prompt for confirmation ---"
235	printf "\n -X # of processors (on parallel hosts) 1"
236	printf "\n "
237	printf "\n The only possible positional parameter is <jobfile>:"
238	printf "\n The complete NQS-job must be provided here."
239	printf "\n <jobfile>=? creates this outline\n\n") \| more
240	exit
241	fi
242
243
244
245	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
246	if [[ "$file_to_send" = "" ]]
247	then
248	printf "\n +++ job-file missing"
249	locat=parameter; exit
250	else
251	if [[ -f $file_to_send ]]
252	then
253	true
254	else
255	printf "\n +++ job-file: "
256	printf "\n $file_to_send"
257	printf "\n does not exist"
258	locat=parameter; exit
259	fi
260	fi
261
262
263
264	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
265	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
266	if [[ $job_name = none ]]
267	then
268	job_name=$file_to_send
269	fi
270	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
271	then
272	printf "\n +++ job-file name: "
273	printf "\n $job_name"
274	printf "\n must not contain \"/\"-characters"
275	locat=parameter; exit
276	fi
277
278
279
280
281	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
282	# ODER WENN HOST NICHT ANGEGEBEN WURDE
283	if [[ $remote_host = none ]]
284	then
285	printf "\n +++ host missing"
286	locat=option; exit
287	else
288	case $remote_host in
289	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
290	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
291	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
292	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
293	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
294	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
295	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
296	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
297	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
298	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
299	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
300	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
301	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
302	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
303	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
304	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
305	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
306	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
307	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
308	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
309	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
310	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
311	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
312	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
313	locat=parameter; exit;;
314	esac
315	fi
316
317
318	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
319	if [[ $no_default_queue != none ]]
320	then
321	error=false
322	ndq=$no_default_queue
323	case $remote_host in
324	(ibm) case $ndq in
325	(p690_express\|p690_standard\|p690_long) error=false;;
326	(*) error=true;;
327	esac;;
328	(ibmh) case $ndq in
329	(no_class) error=false;;
330	(*) error=true;;
331	esac;;
332	(ibmkisti) case $ndq in
333	(class.32plus\|class.1-2\|class.2-32) error=false;;
334	(*) error=true;;
335	esac;;
336	(ibmku) case $ndq in
337	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
338	(*) error=true;;
339	esac;;
340	(ibms) case $ndq in
341	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
342	(*) error=true;;
343	esac;;
344	(ibmy) case $ndq in
345	(parallel) error=false;;
346	(*) error=true;;
347	esac;;
348	(lckiaps) case $ndq in
349	(express\|normal) error=false;;
350	(*) error=true;;
351	esac;;
352	(lckyoto) case $ndq in
353	(eh\|ph) error=false;;
354	(*) error=true;;
355	esac;;
356	(lckyut) case $ndq in
357	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
358	(*) error=true;;
359	esac;;
360	(lcsgib\|lcsgih) case $ndq in
361	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
362	(*) error=true;;
363	esac;;
364	(lctit) case $ndq in
365	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
366	(*) error=true;;
367	esac;;
368	(t3eb) case $ndq in
369	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
370	(*) error=true;;
371	esac;;
372	(necriam) case $ndq in
373	(SP\|SS\|P6) error=false;;
374	(*) error=true;;
375	esac;;
376	(t3eh) case $ndq in
377	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
378	(*) error=true;;
379	esac;;
380	(t3ej2\|t3ej5) case $ndq in
381	(low\|normal\|high) error=false;;
382	(*) error=true;;
383	esac;;
384	(t3es) case $ndq in
385	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
386	(*) error=true;;
387	esac;;
388	(unics) case $ndq in
389	(unics\|ht) error=false;;
390	(*) error=true;;
391	esac;;
392	esac
393	if [[ $error = true ]]
394	then
395	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
396	locat=parameter; exit
397	else
398	queue=$no_default_queue
399	fi
400	fi
401
402
403
404	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
405	# AUFTEILEN
406	done=false
407	while [[ $done = false ]]
408	do
409	if (( $cputime <= 0 ))
410	then
411	printf "\n +++ wrong cpu-time or cpu-time missing"
412	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
413	printf "\n >>> "
414	read cputime 1>/dev/null 2>&1
415	else
416	done=true
417	fi
418	done
419	if [[ $remote_host = nech ]]
420	then
421	if (( tasks_per_node != 0 ))
422	then
423	(( cputime = cputime * tasks_per_node ))
424	elif [[ $numprocs != 0 ]]
425	then
426	(( cputime = cputime * numprocs ))
427	fi
428	fi
429	(( stunden = cputime / 3600 ))
430	(( resttime = cputime - stunden * 3600 ))
431	(( minuten = resttime / 60 ))
432	(( sekunden = resttime - minuten * 60 ))
433	timestring=${stunden}:${minuten}:${sekunden}
434
435
436
437	# PRUEFEN DER KERNSPEICHERANFORDERUNG
438	done=false
439	while [[ $done = false ]]
440	do
441	if (( memory <= 0 ))
442	then
443	printf "\n +++ wrong memory demand or memory demand missing"
444	printf "\n >>> Please type memory in MByte per process as INTEGER:"
445	printf "\n >>> "
446	read memory 1>/dev/null 2>&1
447	else
448	done=true
449	fi
450	done
451
452	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
453	then
454	if (( tasks_per_node != 0 ))
455	then
456	(( Memory = memory * tasks_per_node / 1000 ))
457	elif [[ $numprocs != 0 ]]
458	then
459	(( Memory = memory * numprocs / 1000 ))
460	else
461	(( Memory = memory / 1000 ))
462	fi
463	elif [[ $remote_host = lctit ]]
464	then
465	(( Memory = memory * tasks_per_node / 1000 ))
466	fi
467
468
469	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
470	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
471	then
472	(( memory = memory * threads_per_task ))
473	fi
474
475
476	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
477	if (( tasks_per_node != 0 ))
478	then
479	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
480	fi
481
482
483	# Calculate number of processes per node
484	(( processes_per_node = tasks_per_node * threads_per_task ))
485
486
487	# Calculate number of MPI tasks
488	(( mpi_tasks = numprocs / threads_per_task ))
489
490
491	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
492	if [[ "$scp_port" != "" ]]
493	then
494	PORTOPT="-P $scp_port"
495	SSH_PORTOPT="-p $scp_port"
496	fi
497
498
499	# HEADER-AUSGABE
500	if [[ $verify = true ]]
501	then
502	printf "\n\n"
503	printf "#--------------------------------------------------------------# \n"
504	spalte1=SUBJOB;spalte2=$(date)
505	printf "\| $spalte1$spalte2 \| \n"
506	printf "\| \| \n"
507	printf "\| values of parameters/options: \| \n"
508	spalte1=local_host$punkte; spalte2=$punkte$local_host
509	printf "\| $spalte1$spalte2 \| \n"
510	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
511	printf "\| $spalte1$spalte2 \| \n"
512	spalte1=queue$punkte; spalte2=$punkte$queue
513	printf "\| $spalte1$spalte2 \| \n"
514	spalte1=memory$punkte; spalte2="$punkte$memory mb"
515	printf "\| $spalte1$spalte2 \| \n"
516	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
517	printf "\| $spalte1$spalte2 \| \n"
518	spalte1=job_name$punkte; spalte2="$punkte$job_name"
519	printf "\| $spalte1$spalte2 \| \n"
520	printf "#--------------------------------------------------------------# \n\n"
521
522
523	# KONTROLLABFRAGE, OB ALLES O.K.
524	antwort="dummy"
525	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
526	do
527	read antwort?" >>> continue (y/n) ? "
528	done
529	if [[ $antwort = n \|\| $antwort = N ]]
530	then
531	locat=verify; exit
532	fi
533	printf "\n"
534	fi
535
536
537
538	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
539	kennung=$RANDOM
540	job_on_remhost=${job_name}_${kennung}_$local_host
541	job_to_send=job_to_send_$kennung
542	if [[ $delete_dayfile = false ]]
543	then
544	remote_dayfile=${local_host}_${job_name}_result_$kennung
545	local_dayfile=${remote_host}_${job_name}
546	else
547	remote_dayfile=/dev/null
548	fi
549
550
551
552	# Generate the batch job scripts (qsub/msub/LoadLeveler)
553	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
554	then
555
556	# General LoadLeveler settings
557	execute_in_shell="#!/bin/ksh"
558	use_shell="# @ shell = /bin/ksh"
559	consumable_memory="ConsumableMemory($memory mb)"
560	class="# @ class = $queue"
561	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
562	network_to_use="# @ network.mpi = sn_all,shared,us"
563	data_limit="# @ data_limit = 1.76gb"
564	image_size="# @ image_size = 50"
565	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
566
567	if [[ $email_notification = none ]]
568	then
569	notify_user=""
570	else
571	notify_user="# @ notify_user = $email_notification"
572	if [[ $delete_dayfile = true ]]
573	then
574	notification='# @ notification = never'
575	fi
576	fi
577
578	if [[ $remote_host = ibmh ]]
579	then
580	data_limit=""
581	network_to_use=""
582	class=""
583	environment=""
584	rset="# @ rset = RSET_MCM_AFFINITY"
585	task_affinity="# @ task_affinity = core(1)"
586	elif [[ $remote_host = ibmkisti ]]
587	then
588	network_to_use="# @ network.MPI = sn_all,shared,US"
589	wall_clock_limit="# @ wall_clock_limit = $timestring"
590	if [[ $threads_per_task = 1 ]]
591	then
592	rset="# @ rset = RSET_MCM_AFFINITY"
593	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
594	fi
595	environment=""
596	use_shell=""
597	data_limit=""
598	image_size=""
599	elif [[ $remote_host = ibmku ]]
600	then
601	execute_in_shell="#!/usr/bin/ksh"
602	use_shell="# @ shell = /usr/bin/ksh"
603	consumable_memory=""
604	environment=""
605	network_to_use="# @ network.mpi = sn_all,shared,us"
606	data_limit=""
607	image_size=""
608	elif [[ $remote_host = ibms ]]
609	then
610	network_to_use="# @ network.mpi = csss,shared,us"
611	elif [[ $remote_host = ibmy ]]
612	then
613	consumable_memory=""
614	network_to_use=""
615	fi
616
617	cat > $job_to_send << %%END%%
618	$execute_in_shell
619	$use_shell
620
621	# @ job_type = parallel
622	# @ job_name = $job_name
623	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
624	# @ output = $remote_dayfile
625	# @ error = $remote_dayfile
626	$wall_clock_limit
627	$image_size
628	$class
629	$environment
630	$network_to_use
631	$data_limit
632	$rset
633	$mcm_affinity_options
634	$task_affinity
635	$notification
636	$notify_user
637
638	%%END%%
639
640	if (( nodes > 0 ))
641	then
642
643	if [[ $remote_host != ibmkisti ]]
644	then
645
646	cat >> $job_to_send << %%END%%
647	# @ node = $nodes
648	# @ tasks_per_node = $processes_per_node
649	# @ node_usage = $node_usage
650	# @ queue
651
652	%%END%%
653
654	else
655
656	cat >> $job_to_send << %%END%%
657	# @ total_tasks = $mpi_tasks
658	# @ blocking = unlimited
659	# @ queue
660
661	%%END%%
662
663	fi
664
665	else
666
667	if [[ $remote_host != ibmy ]]
668	then
669
670	cat >> $job_to_send << %%END%%
671	# @ blocking = unlimited
672	# @ total_tasks = $numprocs
673	# @ node_usage = $node_usage
674	# @ queue
675
676	%%END%%
677
678	else
679
680	cat >> $job_to_send << %%END%%
681	# @ node = 1
682	# @ total_tasks = $numprocs
683	# @ queue
684
685	%%END%%
686
687	fi
688
689	fi
690
691	# workaround because of silly job filter on ibmkisti
692	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
693	then
694	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
695	fi
696
697	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
698	then
699
700	cat > $job_to_send << %%END%%
701	#!/bin/ksh
702
703	# @ job_type = serial
704	# @ node_usage = $node_usage
705	# @ job_name = palm
706	# @ wall_clock_limit = ${timestring},$timestring
707	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
708	# @ output = $remote_dayfile
709	# @ error = $remote_dayfile
710	$class
711	$notification
712
713	# @ queue
714
715	%%END%%
716
717	elif [[ $remote_host = lcfimm ]]
718	then
719
720	if [[ $numprocs != 0 ]]
721	then
722	cat > $job_to_send << %%END%%
723	#!/bin/ksh
724	#PBS -N $job_name
725	#PBS -A $project_account
726	#PBS -l walltime=$timestring
727	#PBS -l nodes=${nodes}:ppn=$processes_per_node
728	#PBS -l pmem=${memory}mb
729	#PBS -m abe
730	#PBS -o $remote_dayfile
731	#PBS -j oe
732	mpd &
733
734	%%END%%
735
736	else
737	cat > $job_to_send << %%END%%
738	#!/bin/ksh
739	#PBS -N $job_name
740	#PBS -A $project_account
741	#PBS -l walltime=$timestring
742	#PBS -l ncpus=1
743	#PBS -l pmem=${memory}mb
744	#PBS -m abe
745	#PBS -o $remote_dayfile
746	#PBS -j oe
747
748	%%END%%
749
750	fi
751
752	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
753	then
754
755	if [[ $numprocs != 0 ]]
756	then
757	cat > $job_to_send << %%END%%
758	#!/bin/ksh
759	#PBS -N $job_name
760	#PBS -l walltime=$timestring
761	#PBS -l ncpus=$numprocs
762	#PBS -l pmem=${memory}mb
763	#PBS -o $remote_dayfile
764	#PBS -l nodes=$nodes:ppn=${processes_per_node}
765	#PBS -j oe
766
767	mpd &
768
769	%%END%%
770
771	else
772	cat > $job_to_send << %%END%%
773	#!/bin/ksh
774	#PBS -N $job_name
775	#PBS -l walltime=$timestring
776	#PBS -l ncpus=1
777	#PBS -l pmem=${memory}mb
778	#PBS -o $remote_dayfile
779	#PBS -j oe
780
781	%%END%%
782
783	fi
784
785	elif [[ $remote_host = lckiaps ]]
786	then
787
788	if [[ $numprocs != 0 ]]
789	then
790	cat > $job_to_send << %%END%%
791	#!/bin/ksh
792	#PBS -N $job_name
793	#PBS -l walltime=$timestring
794	#PBS -l select=1:ncpus=$numprocs
795	#PBS -l pmem=${memory}mb
796	#PBS -q $queue
797	#PBS -o $remote_dayfile
798	#PBS -j oe
799	#PBS -V
800
801	%%END%%
802
803	else
804	cat > $job_to_send << %%END%%
805	#!/bin/ksh
806	#PBS -N $job_name
807	#PBS -l walltime=$timestring
808	#PBS -l ncpus=1
809	#PBS -l pmem=${memory}mb
810	#PBS -o $remote_dayfile
811	#PBS -j oe
812
813	%%END%%
814
815	fi
816
817	elif [[ $remote_host = lcyon ]]
818	then
819
820	if [[ $numprocs != 0 ]]
821	then
822	cat > $job_to_send << %%END%%
823	#!/bin/ksh
824	#PBS -N $job_name
825	#PBS -l walltime=$timestring
826	#PBS -l ncpus=$numprocs
827	#PBS -l pmem=${memory}mb
828	#PBS -o $remote_dayfile
829	#PBS -j oe
830
831	%%END%%
832
833	else
834	cat > $job_to_send << %%END%%
835	#!/bin/ksh
836	#PBS -N $job_name
837	#PBS -l walltime=$timestring
838	#PBS -l ncpus=1
839	#PBS -l pmem=${memory}mb
840	#PBS -o $remote_dayfile
841	#PBS -j oe
842
843	%%END%%
844
845	fi
846
847	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
848	then
849
850	if [[ "$sgi_feature" != "" ]]
851	then
852	feature_directive="#PBS -l feature=$sgi_feature"
853	else
854	feature_directive=""
855	fi
856
857	if [[ $queue = dataq \|\| $queue = permq ]]
858	then
859	feature_directive="#PBS -l feature=data"
860	fi
861
862	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
863	then
864	queue_directive="#PBS -q $queue"
865	else
866	queue_directive=""
867	fi
868
869	if [[ $email_notification = none ]]
870	then
871	email_directive=""
872	else
873	email_directive="#PBS -M $email_notification"
874	fi
875
876	if [[ $numprocs != 0 ]]
877	then
878	cat > $job_to_send << %%END%%
879	#!/bin/bash
880	#PBS -N $job_name
881	#PBS -l walltime=$timestring
882	#PBS -l nodes=$nodes:ppn=${processes_per_node}
883	#PBS -l naccesspolicy=$node_usage
884	#PBS -o $remote_dayfile
885	#PBS -j oe
886	$feature_directive
887	$queue_directive
888	$email_directive
889
890	eval \`/sw/swdist/bin/modulesinit\`
891	#. /usr/share/modules/init/bash
892	$init_cmds
893	$module_calls
894
895	echo ld_library_path=\$LD_LIBRARY_PATH
896
897	%%END%%
898
899	else
900	cat > $job_to_send << %%END%%
901	#PBS -S /bin/bash
902	#PBS -N $job_name
903	#PBS -l walltime=$timestring
904	#PBS -l ncpus=1
905	#PBS -o $remote_dayfile
906	#PBS -j oe
907	$queue_directive
908	$email_directive
909
910	eval \`/sw/swdist/bin/modulesinit\`
911	#. /usr/share/modules/init/bash
912	$init_cmds
913	$module_calls
914
915	%%END%%
916
917	fi
918
919	elif [[ $remote_host = lcxe6 ]]
920	then
921
922	if [[ $numprocs != 0 ]]
923	then
924	cat > $job_to_send << %%END%%
925	#!/bin/ksh
926	#PBS -S /bin/ksh
927	#PBS -N $job_name
928	#PBS -A $project_account
929	#PBS -j oe
930	#PBS -l walltime=$timestring
931	#PBS -l mppwidth=${numprocs}
932	#PBS -l mppnppn=${processes_per_node}
933	#PBS -m abe
934	#PBS -o $remote_dayfile
935	$email_directive
936
937	$init_cmds
938	$module_calls
939
940	%%END%%
941
942	else
943	cat > $job_to_send << %%END%%
944	#!/bin/ksh
945	#PBS -S /bin/ksh
946	#PBS -N $job_name
947	#PBS -A $project_account
948	#PBS -j oe
949	#PBS -l walltime=$timestring
950	#PBS -l ncpus=1
951	#PBS -l pmem=${memory}mb
952	#PBS -m abe
953	$email_directive
954	#PBS -o $remote_dayfile
955
956	$init_cmds
957	$module_calls
958
959	%%END%%
960
961	fi
962
963	elif [[ $remote_host = lckyoto ]]
964	then
965
966	cat > $job_to_send << %%END%%
967	#!/bin/ksh
968	# @\$-o $remote_dayfile
969	# @\$-eo -oi
970	# @\$-lP 16
971	# @\$-lp 1
972	# @\$-lm 28gb -llm unlimited -ls unlimited
973	# @\$-q $queue
974	# @\$-Pvn abs_pack
975	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
976	#. /thin/local/etc/setprofile/intel-11.0.sh
977	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
978	. ~/.myprofile
979	#. /home2/t/t51254/palm/current_version/myprofile
980	#. /thin/apps/pgi/mpi.sh
981	#
982	env
983	#
984	set -x
985
986	%%END%%
987
988	elif [[ $remote_host = lcxt5m ]]
989	then
990
991	if [[ $numprocs != 0 ]]
992	then
993	cat > $job_to_send << %%END%%
994	#!/bin/ksh
995	#PBS -S /bin/ksh
996	#PBS -N $job_name
997	#PBS -j oe
998	#PBS -l walltime=$timestring
999	#PBS -l mppwidth=${numprocs}
1000	#PBS -l mppnppn=${processes_per_node}
1001	#PBS -m abe
1002	#PBS -o $remote_dayfile
1003
1004	$init_cmds
1005	$module_calls
1006
1007	%%END%%
1008
1009	else
1010	cat > $job_to_send << %%END%%
1011	#!/bin/ksh
1012	#PBS -S /bin/ksh
1013	#PBS -N $job_name
1014	#PBS -j oe
1015	#PBS -l walltime=$timestring
1016	#PBS -l ncpus=1
1017	#PBS -l pmem=${memory}mb
1018	#PBS -m abe
1019	#PBS -o $remote_dayfile
1020
1021	$init_cmds
1022	$module_calls
1023
1024	%%END%%
1025
1026	fi
1027
1028	elif [[ $remote_host = lckyut ]]
1029	then
1030	cat > $job_to_send << %%END%%
1031	#!/bin/bash
1032	#PJM -L "vnode=$numprocs"
1033	#PJM -L "vnode-core=1"
1034	#PJM -L "rscgrp=$queue"
1035	#PJM --no-stging
1036	#PJM -L "elapse=$timestring"
1037	#PJM --mpi proc=$numprocs
1038	#PJM -o $remote_dayfile
1039	#PJM -j
1040
1041	export LANG=en_US.UTF-8
1042	%%END%%
1043
1044	elif [[ $remote_host = nech ]]
1045	then
1046
1047	if (( nodes > 1 ))
1048	then
1049	cat > $job_to_send << %%END%%
1050	#!/bin/ksh
1051	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1052	#PBS -l ${qsubmem}=${Memory}gb
1053	#PBS -b $nodes
1054	#PBS -o $remote_dayfile
1055	#PBS -N palm
1056	#PBS -j o
1057	#PBS -T mpisx
1058
1059	%%END%%
1060
1061	elif [[ $numprocs != 0 ]]
1062	then
1063	cat > $job_to_send << %%END%%
1064	#!/bin/ksh
1065	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1066	#PBS -l ${qsubmem}=${Memory}gb
1067	#PBS -o $remote_dayfile
1068	#PBS -N palm
1069	#PBS -j o
1070
1071	%%END%%
1072
1073	else
1074	cat > $job_to_send << %%END%%
1075	#!/bin/ksh
1076	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1077	#PBS -o $remote_dayfile
1078	#PBS -j o
1079
1080	%%END%%
1081
1082	fi
1083
1084	elif [[ $remote_host = necriam ]]
1085	then
1086
1087	if (( nodes > 1 ))
1088	then
1089	cat > $job_to_send << %%END%%
1090	#!/bin/ksh
1091	#PBS -b $nodes
1092	#PBS -o $remote_dayfile
1093	#PBS -N $job_name
1094	#PBS -j o
1095	#PBS -v MPIPROGINV=YES
1096
1097	%%END%%
1098
1099	elif [[ $numprocs != 0 ]]
1100	then
1101	cat > $job_to_send << %%END%%
1102	#!/bin/ksh
1103	#PBS -q ${queue}
1104	#PBS -o $remote_dayfile
1105	#PBS -N $job_name
1106	#PBS -j o
1107	#PBS -v MPIPROGINV=YES
1108
1109	%%END%%
1110
1111	else
1112	cat > $job_to_send << %%END%%
1113	#!/bin/ksh
1114	#PBS -o $remote_dayfile
1115	#PBS -j o
1116
1117	%%END%%
1118
1119	fi
1120
1121	elif [[ $remote_host = lctit ]]
1122	then
1123	cat > $job_to_send << %%END%%
1124	#!/bin/ksh
1125	$init_cmds
1126	$module_calls
1127
1128	%%END%%
1129
1130	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1131	if [[ $tasks_per_node != $processes_per_node ]]
1132	then
1133	submcom="$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 $queue "
1134	else
1135	submcom="$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 "
1136	fi
1137
1138	else
1139
1140	cat > $job_to_send << %%END%%
1141	# @\$-q ${queue}
1142	# @\$-l${qsubtime} $timestring
1143	# @\$-l${qsubmem} ${memory}mb
1144	# @\$-o $remote_dayfile
1145	# @\$-eo
1146
1147	%%END%%
1148
1149	fi
1150
1151
1152	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1153	# BEI EXIT VERANLASSEN
1154	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1155	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1156	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1157	then
1158	echo "set +vx" >> $job_to_send
1159	echo "trap '" >> $job_to_send
1160	echo "set +vx" >> $job_to_send
1161	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 \|\| $remote_host = lckyut ]]
1162	then
1163	if [[ $remote_host = ibmh ]]
1164	then
1165	return_queue=c1
1166	elif [[ $remote_host = ibmkisti ]]
1167	then
1168	return_queue=class.1-2
1169	elif [[ $remote_host = ibmku ]]
1170	then
1171	return_queue=sdbg2
1172	elif [[ $remote_host = ibms ]]
1173	then
1174	return_queue=p_normal
1175	elif [[ $remote_host = ibmy ]]
1176	then
1177	return_queue=serial
1178	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1179	then
1180	return_queue=serialq
1181	elif [[ $remote_host = necriam ]]
1182	then
1183	return_queue=SP
1184	elif [[ $remote_host = lckiaps ]]
1185	then
1186	return_queue=express
1187	elif [[ $remote_host = lckyut ]]
1188	then
1189	return_queue=cx-single
1190	else
1191	return_queue=unknown
1192	fi
1193
1194	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1195	then
1196
1197	if [[ $remote_host = ibmku ]]
1198	then
1199	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1200	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1201	else
1202	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1203	fi
1204	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1205	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1206	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1207	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1208	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1209	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1210	if [[ $host != "ibmh" ]]
1211	then
1212	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1213	fi
1214	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1215	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1216
1217	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1218	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1219
1220	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1221	echo "echo \"batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
1222	if [[ $remote_host = ibmku ]]
1223	then
1224	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1225	fi
1226	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1227
1228	elif [[ $remote_host = nech ]]
1229	then
1230	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1231	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1232	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1233	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1234	echo "#PBS -j o" >> $job_to_send
1235	echo " " >> $job_to_send
1236	echo "set -x" >> $job_to_send
1237	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1238	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1239	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1240	echo "%%END%%" >> $job_to_send
1241
1242	elif [[ $remote_host = necriam ]]
1243	then
1244	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1245	echo "#PBS -q $return_queue" >> $job_to_send
1246	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1247	echo "#PBS -j o" >> $job_to_send
1248	echo " " >> $job_to_send
1249	echo "set -x" >> $job_to_send
1250	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1251	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1252	echo "%%END%%" >> $job_to_send
1253
1254	elif [[ $remote_host = lckyut ]]
1255	then
1256	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1257	echo "#!/bin/bash" >> $job_to_send
1258	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1259	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1260	echo "#PJM --no-stging" >> $job_to_send
1261	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1262	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1263	echo "#PJM -j" >> $job_to_send
1264	echo " " >> $job_to_send
1265	echo "export LANG=en_US.UTF-8" >> $job_to_send
1266	echo "set -x" >> $job_to_send
1267	echo "batch_scp $PORTOPT -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
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
1276	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1277	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1278	echo "#PBS -l feature=data" >> $job_to_send
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 $PORTOPT -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%%" >> $job_to_send
1287
1288	else
1289
1290	echo "cat > scpjob.$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 $PORTOPT -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
1317	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1318	then
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
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	elif [[ $remote_host = lckiaps ]]
1329	then
1330	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1331	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1332	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1333	elif [[ $remote_host = lckyut ]]
1334	then
1335	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1336	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1337	else
1338	echo "$submcom scpjob.$kennung" >> $job_to_send
1339	fi
1340	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1341	then
1342	echo "rm scpjob.$kennung" >> $job_to_send
1343	fi
1344	if [[ $remote_host = nech ]]
1345	then
1346	echo "cd -" >> $job_to_send
1347	fi
1348	else
1349	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1350	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1351	fi
1352	echo "set -x" >> $job_to_send
1353	echo " ' exit" >> $job_to_send
1354	echo "set -x" >> $job_to_send
1355	fi
1356
1357
1358
1359	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1360	cat $file_to_send >> $job_to_send
1361
1362	if [[ $remote_host = ibm ]]
1363	then
1364	echo " " >> $job_to_send
1365	echo "exit" >> $job_to_send
1366	fi
1367
1368	# remove job file
1369	if [[ $remote_host = lctit \|\| $remote_host = ibmku ]]
1370	then
1371	echo " " >> $job_to_send
1372	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1373	fi
1374
1375
1376
1377	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1378	if [[ -z $remote_user ]]
1379	then
1380	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1381	then
1382	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1383	fi
1384	fi
1385
1386
1387
1388	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1389	if [[ $no_submit = false ]]
1390	then
1391	if [[ $remote_host != $local_host ]]
1392	then
1393	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1394	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1395	then
1396	job_catalog_save=$job_catalog
1397	job_catalog=job_queue
1398	elif [[ $remote_host = nech ]]
1399	then
1400	job_catalog_save=$job_catalog
1401	job_catalog=/hpf/b/${remote_user}/job_queue
1402	fi
1403	if [[ $local_host = decalpha ]]
1404	then
1405	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1406	# PASSWORT AB
1407	/bin/scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1408	elif [[ $remote_host = nech ]]
1409	then
1410	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1411	# TRANSFERIERT WERDEN
1412	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1413	else
1414	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1415	fi
1416	if [[ $? = 1 ]]
1417	then
1418	locat=scp; exit
1419	fi
1420	if [[ $remote_host = ibms ]]
1421	then
1422	job_catalog=$job_catalog_save
1423	fi
1424	[[ $verify = true ]] && printf "\n >>> finished\n"
1425	else
1426	eval job_catalog=$job_catalog
1427	cp $job_to_send ${job_catalog}/$job_on_remhost
1428	fi
1429
1430
1431
1432	# NQS- BZW. LOADLEVELER-JOB STARTEN
1433	if [[ $remote_host != $local_host ]]
1434	then
1435	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1436
1437	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1438	then
1439	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1440	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1441	elif [[ $remote_host = ibmku ]]
1442	then
1443	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1444	else
1445	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1446	fi
1447
1448	[[ $verify = true ]] && printf " >>> o.k.\n"
1449	else
1450	cd $job_catalog
1451	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1452	then
1453	eval $submcom $job_on_remhost
1454	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyut ]]
1455	then
1456	chmod u+x $job_on_remhost
1457	echo "$submcom $job_on_remhost"
1458	eval $submcom $job_on_remhost
1459	elif [[ $local_host = nech ]]
1460	then
1461	if [[ $queue = default ]]
1462	then
1463	eval $submcom $job_on_remhost
1464	else
1465	eval $submcom -q $queue $job_on_remhost
1466	fi
1467	else
1468	qsub $job_on_remhost
1469	fi
1470
1471	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1472	# only at the end of the job.
1473	if [[ $local_host != lctit && $local_host != ibmku ]]
1474	then
1475	rm $job_on_remhost
1476	fi
1477	cd - > /dev/null
1478	fi
1479	fi
1480
1481
1482
1483	# ABSCHLUSSARBEITEN
1484
1485	if [[ $no_submit = false ]]
1486	then
1487	rm $job_to_send
1488	fi
1489	[[ $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 |