Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1096

Last change on this file since 1096 was 1096, checked in by raasch, 11 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 47.9 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	#
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1096 2013-02-03 02:49:41Z raasch $
30	#
31	# 1094 2013-02-03 01:52:12Z raasch
32	# new option -P for explicit setting of ssh/scp port,
33	# decalpha parts (yonsei) removed
34	#
35	# 2013-02-02 07:06:13Z raasch
36	# adjustments for Kyushu-University computing center (lckyut)
37	# old changelog messages removed
38	#
39	# 1046 2012-11-09 14:38:45Z maronga
40	# code put under GPL (PALM 3.9)
41	#
42	# 08/07/94 - Siggi - first version finished
43	# 29/06/94 - Siggi - script development started
44	#--------------------------------------------------------------------------------#
45	# subjob - script for automatic generation and submission of batch-job files
46	# for various batch queuing systems
47	#--------------------------------------------------------------------------------#
48
49
50	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
51	delete_dayfile=false
52	email_notification=none
53	group_number=none
54	locat=normal
55	no_default_queue=none
56	no_submit=false
57	job_catalog="~/job_queue"
58	job_name=none
59	local_user=$LOGNAME
60	node_usage=shared
61	numprocs=0
62	punkte="..........................................................."
63	submcom=qsub
64	queue=default
65	remote_host=none
66	remote_user=""
67	verify=true
68
69	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
70	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
71	typeset -L20 spalte1
72	typeset -R40 spalte2
73	typeset -L60 spalte3
74
75
76
77	# FEHLERBEHANDLUNG
78	# BEI EXIT:
79	trap 'if [[ $locat != normal ]]
80	then
81	case $locat in
82	(option) printf "\n --> available optios can be displayed"
83	printf " by typing:"
84	printf "\n \"subjob ?\" \n";;
85	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
86	(*) printf "\n +++ unknown error"
87	printf "\n please inform S. Raasch!\n"
88	esac
89	[[ -f $job_to_send ]] && rm $job_to_send
90	printf "\n\n+++ SUBJOB killed \n\n"
91	fi' exit
92
93
94	# BEI TERMINAL-BREAK:
95	trap '[[ -f $job_to_send ]] && rm $job_to_send
96	printf "\n\n+++ SUBJOB killed \n\n"
97	exit
98	' 2
99
100
101
102
103	# LOKALEN HOSTNAMEN FESTSTELLEN
104	local_host=$(hostname)
105
106
107
108	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
109	# UEBERHAUPT ZULAESSIG IST
110	# Note: One of the entries for "lck" or "lckordi" always has to be
111	# comment out, because the hostname (node*) is same for both machines
112	case $local_host in
113	(atmos) local_addres=172.20.25.35; local_host=lcide;;
114	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
115	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
116	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
117	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
118	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
119	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
120	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
121	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
122	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
123	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
124	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
125	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
126	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
127	(cs*) local_addres=136.172.44.131; local_host=nech;;
128	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
129	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
130	(node*) local_addres=165.132.26.61 local_host=lck;;
131	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
132	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
133	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
134	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
135	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
136	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
137	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
138	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
139	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
140	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
141	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
142	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
143	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
144	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
145	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
146	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
147	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
148	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
149	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
150	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
151	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
152	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
153	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
154	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
155	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
156	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
157	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
158	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
159	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
160	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
161	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
162	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
163	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
164	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
165	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
166	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
167	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
168	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
169	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
170	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
171	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
172	(*) printf "\n +++ \"$local_host\" unknown";
173	printf "\n please inform S. Raasch!";
174	locat=parameter; exit;;
175	esac
176
177
178
179	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
180	remote_host=$local_host
181
182
183
184
185	# PROZEDUROPTIONEN EINLESEN
186	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
187	do
188	case $option in
189	(c) job_catalog=$OPTARG;;
190	(d) delete_dayfile=true;;
191	(D) no_submit=true;;
192	(e) email_notification=$OPTARG;;
193	(g) group_number=$OPTARG;;
194	(h) remote_host=$OPTARG;;
195	(m) memory=$OPTARG;;
196	(n) job_name=$OPTARG;;
197	(N) node_usage=$OPTARG;;
198	(O) threads_per_task=$OPTARG;;
199	(P) scp_port=$OPTARG;;
200	(q) no_default_queue=$OPTARG;;
201	(t) cputime=$OPTARG;;
202	(T) tasks_per_node=$OPTARG;;
203	(u) remote_user=$OPTARG;;
204	(v) verify=false;;
205	(X) numprocs=$OPTARG;;
206	(\?) printf "\n +++ Option $OPTARG unknown \n";
207	locat=option; exit;;
208	esac
209	done
210
211
212	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
213	shift OPTIND-1; file_to_send=$1
214
215
216	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
217	if [ "$1" = "?" ]
218	then
219	(printf "\n *** subjob can be called as follows:\n"
220	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
221	printf "\n Description of available options:\n"
222	printf "\n Option Description Default-Value"
223	printf "\n -c job-input- and output-catalog ~/job_queue"
224	printf "\n -d no job-protocol will be created ---"
225	printf "\n -D only the job-file will be created ---"
226	printf "\n -h execution host, available hosts: $remote_host"
227	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
228	printf "\n lckiaps, lctit, nech, necriam, unics"
229	printf "\n -m memory demand per process in MByte ---"
230	printf "\n -n jobname <jobdatei>"
231	printf "\n -O threads per task (for OpenMP usage) 1"
232	printf "\n -P ssh/scp port default port"
233	printf "\n -q job-queue to be used default"
234	printf "\n -t allowed cpu-time in seconds ---"
235	printf "\n -T tasks per node (on parallel hosts) ---"
236	printf "\n -u username on execution host from .netrc"
237	printf "\n -v no prompt for confirmation ---"
238	printf "\n -X # of processors (on parallel hosts) 1"
239	printf "\n "
240	printf "\n The only possible positional parameter is <jobfile>:"
241	printf "\n The complete NQS-job must be provided here."
242	printf "\n <jobfile>=? creates this outline\n\n") \| more
243	exit
244	fi
245
246
247
248	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
249	if [[ "$file_to_send" = "" ]]
250	then
251	printf "\n +++ job-file missing"
252	locat=parameter; exit
253	else
254	if [[ -f $file_to_send ]]
255	then
256	true
257	else
258	printf "\n +++ job-file: "
259	printf "\n $file_to_send"
260	printf "\n does not exist"
261	locat=parameter; exit
262	fi
263	fi
264
265
266
267	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
268	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
269	if [[ $job_name = none ]]
270	then
271	job_name=$file_to_send
272	fi
273	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
274	then
275	printf "\n +++ job-file name: "
276	printf "\n $job_name"
277	printf "\n must not contain \"/\"-characters"
278	locat=parameter; exit
279	fi
280
281
282
283
284	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
285	# ODER WENN HOST NICHT ANGEGEBEN WURDE
286	if [[ $remote_host = none ]]
287	then
288	printf "\n +++ host missing"
289	locat=option; exit
290	else
291	case $remote_host in
292	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
293	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
294	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
295	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
296	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
297	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
298	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
299	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
300	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
301	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
302	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
303	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
304	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
305	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
306	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
307	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
308	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
309	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
310	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
311	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
312	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
313	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
314	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
315	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
316	locat=parameter; exit;;
317	esac
318	fi
319
320
321	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
322	if [[ $no_default_queue != none ]]
323	then
324	error=false
325	ndq=$no_default_queue
326	case $remote_host in
327	(ibm) case $ndq in
328	(p690_express\|p690_standard\|p690_long) error=false;;
329	(*) error=true;;
330	esac;;
331	(ibmh) case $ndq in
332	(no_class) error=false;;
333	(*) error=true;;
334	esac;;
335	(ibmkisti) case $ndq in
336	(class.32plus\|class.1-2\|class.2-32) error=false;;
337	(*) error=true;;
338	esac;;
339	(ibmku) case $ndq in
340	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
341	(*) error=true;;
342	esac;;
343	(ibms) case $ndq in
344	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
345	(*) error=true;;
346	esac;;
347	(ibmy) case $ndq in
348	(parallel) error=false;;
349	(*) error=true;;
350	esac;;
351	(lckiaps) case $ndq in
352	(express\|normal) error=false;;
353	(*) error=true;;
354	esac;;
355	(lckyoto) case $ndq in
356	(eh\|ph) error=false;;
357	(*) error=true;;
358	esac;;
359	(lckyut) case $ndq in
360	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
361	(*) error=true;;
362	esac;;
363	(lcsgib\|lcsgih) case $ndq in
364	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
365	(*) error=true;;
366	esac;;
367	(lctit) case $ndq in
368	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
369	(*) error=true;;
370	esac;;
371	(t3eb) case $ndq in
372	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
373	(*) error=true;;
374	esac;;
375	(necriam) case $ndq in
376	(SP\|SS\|P6) error=false;;
377	(*) error=true;;
378	esac;;
379	(t3eh) case $ndq in
380	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
381	(*) error=true;;
382	esac;;
383	(t3ej2\|t3ej5) case $ndq in
384	(low\|normal\|high) error=false;;
385	(*) error=true;;
386	esac;;
387	(t3es) case $ndq in
388	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
389	(*) error=true;;
390	esac;;
391	(unics) case $ndq in
392	(unics\|ht) error=false;;
393	(*) error=true;;
394	esac;;
395	esac
396	if [[ $error = true ]]
397	then
398	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
399	locat=parameter; exit
400	else
401	queue=$no_default_queue
402	fi
403	fi
404
405
406
407	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
408	# AUFTEILEN
409	done=false
410	while [[ $done = false ]]
411	do
412	if (( $cputime <= 0 ))
413	then
414	printf "\n +++ wrong cpu-time or cpu-time missing"
415	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
416	printf "\n >>> "
417	read cputime 1>/dev/null 2>&1
418	else
419	done=true
420	fi
421	done
422	if [[ $remote_host = nech ]]
423	then
424	if (( tasks_per_node != 0 ))
425	then
426	(( cputime = cputime * tasks_per_node ))
427	elif [[ $numprocs != 0 ]]
428	then
429	(( cputime = cputime * numprocs ))
430	fi
431	fi
432	(( stunden = cputime / 3600 ))
433	(( resttime = cputime - stunden * 3600 ))
434	(( minuten = resttime / 60 ))
435	(( sekunden = resttime - minuten * 60 ))
436	timestring=${stunden}:${minuten}:${sekunden}
437
438
439
440	# PRUEFEN DER KERNSPEICHERANFORDERUNG
441	done=false
442	while [[ $done = false ]]
443	do
444	if (( memory <= 0 ))
445	then
446	printf "\n +++ wrong memory demand or memory demand missing"
447	printf "\n >>> Please type memory in MByte per process as INTEGER:"
448	printf "\n >>> "
449	read memory 1>/dev/null 2>&1
450	else
451	done=true
452	fi
453	done
454
455	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
456	then
457	if (( tasks_per_node != 0 ))
458	then
459	(( Memory = memory * tasks_per_node / 1000 ))
460	elif [[ $numprocs != 0 ]]
461	then
462	(( Memory = memory * numprocs / 1000 ))
463	else
464	(( Memory = memory / 1000 ))
465	fi
466	elif [[ $remote_host = lctit ]]
467	then
468	(( Memory = memory * tasks_per_node / 1000 ))
469	fi
470
471
472	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
473	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
474	then
475	(( memory = memory * threads_per_task ))
476	fi
477
478
479	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
480	if (( tasks_per_node != 0 ))
481	then
482	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
483	fi
484
485
486	# Calculate number of processes per node
487	(( processes_per_node = tasks_per_node * threads_per_task ))
488
489
490	# Calculate number of MPI tasks
491	(( mpi_tasks = numprocs / threads_per_task ))
492
493
494	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
495	if [[ "$scp_port" != "" ]]
496	then
497	PORTOPT="-P $scp_port"
498	SSH_PORTOPT="-p $scp_port"
499	fi
500
501
502	# HEADER-AUSGABE
503	if [[ $verify = true ]]
504	then
505	printf "\n\n"
506	printf "#--------------------------------------------------------------# \n"
507	spalte1=SUBJOB;spalte2=$(date)
508	printf "\| $spalte1$spalte2 \| \n"
509	printf "\| \| \n"
510	printf "\| values of parameters/options: \| \n"
511	spalte1=local_host$punkte; spalte2=$punkte$local_host
512	printf "\| $spalte1$spalte2 \| \n"
513	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
514	printf "\| $spalte1$spalte2 \| \n"
515	spalte1=queue$punkte; spalte2=$punkte$queue
516	printf "\| $spalte1$spalte2 \| \n"
517	spalte1=memory$punkte; spalte2="$punkte$memory mb"
518	printf "\| $spalte1$spalte2 \| \n"
519	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
520	printf "\| $spalte1$spalte2 \| \n"
521	spalte1=job_name$punkte; spalte2="$punkte$job_name"
522	printf "\| $spalte1$spalte2 \| \n"
523	printf "#--------------------------------------------------------------# \n\n"
524
525
526	# KONTROLLABFRAGE, OB ALLES O.K.
527	antwort="dummy"
528	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
529	do
530	read antwort?" >>> continue (y/n) ? "
531	done
532	if [[ $antwort = n \|\| $antwort = N ]]
533	then
534	locat=verify; exit
535	fi
536	printf "\n"
537	fi
538
539
540
541	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
542	kennung=$RANDOM
543	job_on_remhost=${job_name}_${kennung}_$local_host
544	job_to_send=job_to_send_$kennung
545	if [[ $delete_dayfile = false ]]
546	then
547	remote_dayfile=${local_host}_${job_name}_result_$kennung
548	local_dayfile=${remote_host}_${job_name}
549	else
550	remote_dayfile=/dev/null
551	fi
552
553
554
555	# Generate the batch job scripts (qsub/msub/LoadLeveler)
556	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
557	then
558
559	# General LoadLeveler settings
560	execute_in_shell="#!/bin/ksh"
561	use_shell="# @ shell = /bin/ksh"
562	consumable_memory="ConsumableMemory($memory mb)"
563	class="# @ class = $queue"
564	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
565	network_to_use="# @ network.mpi = sn_all,shared,us"
566	data_limit="# @ data_limit = 1.76gb"
567	image_size="# @ image_size = 50"
568	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
569
570	if [[ $email_notification = none ]]
571	then
572	notify_user=""
573	else
574	notify_user="# @ notify_user = $email_notification"
575	if [[ $delete_dayfile = true ]]
576	then
577	notification='# @ notification = never'
578	fi
579	fi
580
581	if [[ $remote_host = ibmh ]]
582	then
583	data_limit=""
584	network_to_use=""
585	class=""
586	environment=""
587	rset="# @ rset = RSET_MCM_AFFINITY"
588	task_affinity="# @ task_affinity = core(1)"
589	elif [[ $remote_host = ibmkisti ]]
590	then
591	network_to_use="# @ network.MPI = sn_all,shared,US"
592	wall_clock_limit="# @ wall_clock_limit = $timestring"
593	if [[ $threads_per_task = 1 ]]
594	then
595	rset="# @ rset = RSET_MCM_AFFINITY"
596	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
597	fi
598	environment=""
599	use_shell=""
600	data_limit=""
601	image_size=""
602	elif [[ $remote_host = ibmku ]]
603	then
604	execute_in_shell="#!/usr/bin/ksh"
605	use_shell="# @ shell = /usr/bin/ksh"
606	consumable_memory=""
607	environment=""
608	network_to_use="# @ network.mpi = sn_all,shared,us"
609	data_limit=""
610	image_size=""
611	elif [[ $remote_host = ibms ]]
612	then
613	network_to_use="# @ network.mpi = csss,shared,us"
614	elif [[ $remote_host = ibmy ]]
615	then
616	consumable_memory=""
617	network_to_use=""
618	fi
619
620	cat > $job_to_send << %%END%%
621	$execute_in_shell
622	$use_shell
623
624	# @ job_type = parallel
625	# @ job_name = $job_name
626	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
627	# @ output = $remote_dayfile
628	# @ error = $remote_dayfile
629	$wall_clock_limit
630	$image_size
631	$class
632	$environment
633	$network_to_use
634	$data_limit
635	$rset
636	$mcm_affinity_options
637	$task_affinity
638	$notification
639	$notify_user
640
641	%%END%%
642
643	if (( nodes > 0 ))
644	then
645
646	if [[ $remote_host != ibmkisti ]]
647	then
648
649	cat >> $job_to_send << %%END%%
650	# @ node = $nodes
651	# @ tasks_per_node = $processes_per_node
652	# @ node_usage = $node_usage
653	# @ queue
654
655	%%END%%
656
657	else
658
659	cat >> $job_to_send << %%END%%
660	# @ total_tasks = $mpi_tasks
661	# @ blocking = unlimited
662	# @ queue
663
664	%%END%%
665
666	fi
667
668	else
669
670	if [[ $remote_host != ibmy ]]
671	then
672
673	cat >> $job_to_send << %%END%%
674	# @ blocking = unlimited
675	# @ total_tasks = $numprocs
676	# @ node_usage = $node_usage
677	# @ queue
678
679	%%END%%
680
681	else
682
683	cat >> $job_to_send << %%END%%
684	# @ node = 1
685	# @ total_tasks = $numprocs
686	# @ queue
687
688	%%END%%
689
690	fi
691
692	fi
693
694	# workaround because of silly job filter on ibmkisti
695	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
696	then
697	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
698	fi
699
700	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
701	then
702
703	cat > $job_to_send << %%END%%
704	#!/bin/ksh
705
706	# @ job_type = serial
707	# @ node_usage = $node_usage
708	# @ job_name = palm
709	# @ wall_clock_limit = ${timestring},$timestring
710	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
711	# @ output = $remote_dayfile
712	# @ error = $remote_dayfile
713	$class
714	$notification
715
716	# @ queue
717
718	%%END%%
719
720	elif [[ $remote_host = lcfimm ]]
721	then
722
723	if [[ $numprocs != 0 ]]
724	then
725	cat > $job_to_send << %%END%%
726	#!/bin/ksh
727	#PBS -N $job_name
728	#PBS -A $project_account
729	#PBS -l walltime=$timestring
730	#PBS -l nodes=${nodes}:ppn=$processes_per_node
731	#PBS -l pmem=${memory}mb
732	#PBS -m abe
733	#PBS -o $remote_dayfile
734	#PBS -j oe
735	mpd &
736
737	%%END%%
738
739	else
740	cat > $job_to_send << %%END%%
741	#!/bin/ksh
742	#PBS -N $job_name
743	#PBS -A $project_account
744	#PBS -l walltime=$timestring
745	#PBS -l ncpus=1
746	#PBS -l pmem=${memory}mb
747	#PBS -m abe
748	#PBS -o $remote_dayfile
749	#PBS -j oe
750
751	%%END%%
752
753	fi
754
755	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
756	then
757
758	if [[ $numprocs != 0 ]]
759	then
760	cat > $job_to_send << %%END%%
761	#!/bin/ksh
762	#PBS -N $job_name
763	#PBS -l walltime=$timestring
764	#PBS -l ncpus=$numprocs
765	#PBS -l pmem=${memory}mb
766	#PBS -o $remote_dayfile
767	#PBS -l nodes=$nodes:ppn=${processes_per_node}
768	#PBS -j oe
769
770	mpd &
771
772	%%END%%
773
774	else
775	cat > $job_to_send << %%END%%
776	#!/bin/ksh
777	#PBS -N $job_name
778	#PBS -l walltime=$timestring
779	#PBS -l ncpus=1
780	#PBS -l pmem=${memory}mb
781	#PBS -o $remote_dayfile
782	#PBS -j oe
783
784	%%END%%
785
786	fi
787
788	elif [[ $remote_host = lckiaps ]]
789	then
790
791	if [[ $numprocs != 0 ]]
792	then
793	cat > $job_to_send << %%END%%
794	#!/bin/ksh
795	#PBS -N $job_name
796	#PBS -l walltime=$timestring
797	#PBS -l select=1:ncpus=$numprocs
798	#PBS -l pmem=${memory}mb
799	#PBS -q $queue
800	#PBS -o $remote_dayfile
801	#PBS -j oe
802	#PBS -V
803
804	%%END%%
805
806	else
807	cat > $job_to_send << %%END%%
808	#!/bin/ksh
809	#PBS -N $job_name
810	#PBS -l walltime=$timestring
811	#PBS -l ncpus=1
812	#PBS -l pmem=${memory}mb
813	#PBS -o $remote_dayfile
814	#PBS -j oe
815
816	%%END%%
817
818	fi
819
820	elif [[ $remote_host = lcyon ]]
821	then
822
823	if [[ $numprocs != 0 ]]
824	then
825	cat > $job_to_send << %%END%%
826	#!/bin/ksh
827	#PBS -N $job_name
828	#PBS -l walltime=$timestring
829	#PBS -l ncpus=$numprocs
830	#PBS -l pmem=${memory}mb
831	#PBS -o $remote_dayfile
832	#PBS -j oe
833
834	%%END%%
835
836	else
837	cat > $job_to_send << %%END%%
838	#!/bin/ksh
839	#PBS -N $job_name
840	#PBS -l walltime=$timestring
841	#PBS -l ncpus=1
842	#PBS -l pmem=${memory}mb
843	#PBS -o $remote_dayfile
844	#PBS -j oe
845
846	%%END%%
847
848	fi
849
850	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
851	then
852
853	if [[ "$sgi_feature" != "" ]]
854	then
855	feature_directive="#PBS -l feature=$sgi_feature"
856	else
857	feature_directive=""
858	fi
859
860	if [[ $queue = dataq \|\| $queue = permq ]]
861	then
862	feature_directive="#PBS -l feature=data"
863	fi
864
865	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
866	then
867	queue_directive="#PBS -q $queue"
868	else
869	queue_directive=""
870	fi
871
872	if [[ $email_notification = none ]]
873	then
874	email_directive=""
875	else
876	email_directive="#PBS -M $email_notification"
877	fi
878
879	if [[ $numprocs != 0 ]]
880	then
881	cat > $job_to_send << %%END%%
882	#!/bin/bash
883	#PBS -N $job_name
884	#PBS -l walltime=$timestring
885	#PBS -l nodes=$nodes:ppn=${processes_per_node}
886	#PBS -l naccesspolicy=$node_usage
887	#PBS -o $remote_dayfile
888	#PBS -j oe
889	$feature_directive
890	$queue_directive
891	$email_directive
892
893	eval \`/sw/swdist/bin/modulesinit\`
894	#. /usr/share/modules/init/bash
895	$init_cmds
896	$module_calls
897
898	echo ld_library_path=\$LD_LIBRARY_PATH
899
900	%%END%%
901
902	else
903	cat > $job_to_send << %%END%%
904	#PBS -S /bin/bash
905	#PBS -N $job_name
906	#PBS -l walltime=$timestring
907	#PBS -l ncpus=1
908	#PBS -o $remote_dayfile
909	#PBS -j oe
910	$queue_directive
911	$email_directive
912
913	eval \`/sw/swdist/bin/modulesinit\`
914	#. /usr/share/modules/init/bash
915	$init_cmds
916	$module_calls
917
918	%%END%%
919
920	fi
921
922	elif [[ $remote_host = lcxe6 ]]
923	then
924
925	if [[ $numprocs != 0 ]]
926	then
927	cat > $job_to_send << %%END%%
928	#!/bin/ksh
929	#PBS -S /bin/ksh
930	#PBS -N $job_name
931	#PBS -A $project_account
932	#PBS -j oe
933	#PBS -l walltime=$timestring
934	#PBS -l mppwidth=${numprocs}
935	#PBS -l mppnppn=${processes_per_node}
936	#PBS -m abe
937	#PBS -o $remote_dayfile
938	$email_directive
939
940	$init_cmds
941	$module_calls
942
943	%%END%%
944
945	else
946	cat > $job_to_send << %%END%%
947	#!/bin/ksh
948	#PBS -S /bin/ksh
949	#PBS -N $job_name
950	#PBS -A $project_account
951	#PBS -j oe
952	#PBS -l walltime=$timestring
953	#PBS -l ncpus=1
954	#PBS -l pmem=${memory}mb
955	#PBS -m abe
956	$email_directive
957	#PBS -o $remote_dayfile
958
959	$init_cmds
960	$module_calls
961
962	%%END%%
963
964	fi
965
966	elif [[ $remote_host = lckyoto ]]
967	then
968
969	cat > $job_to_send << %%END%%
970	#!/bin/ksh
971	# @\$-o $remote_dayfile
972	# @\$-eo -oi
973	# @\$-lP 16
974	# @\$-lp 1
975	# @\$-lm 28gb -llm unlimited -ls unlimited
976	# @\$-q $queue
977	# @\$-Pvn abs_pack
978	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
979	#. /thin/local/etc/setprofile/intel-11.0.sh
980	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
981	. ~/.myprofile
982	#. /home2/t/t51254/palm/current_version/myprofile
983	#. /thin/apps/pgi/mpi.sh
984	#
985	env
986	#
987	set -x
988
989	%%END%%
990
991	elif [[ $remote_host = lcxt5m ]]
992	then
993
994	if [[ $numprocs != 0 ]]
995	then
996	cat > $job_to_send << %%END%%
997	#!/bin/ksh
998	#PBS -S /bin/ksh
999	#PBS -N $job_name
1000	#PBS -j oe
1001	#PBS -l walltime=$timestring
1002	#PBS -l mppwidth=${numprocs}
1003	#PBS -l mppnppn=${processes_per_node}
1004	#PBS -m abe
1005	#PBS -o $remote_dayfile
1006
1007	$init_cmds
1008	$module_calls
1009
1010	%%END%%
1011
1012	else
1013	cat > $job_to_send << %%END%%
1014	#!/bin/ksh
1015	#PBS -S /bin/ksh
1016	#PBS -N $job_name
1017	#PBS -j oe
1018	#PBS -l walltime=$timestring
1019	#PBS -l ncpus=1
1020	#PBS -l pmem=${memory}mb
1021	#PBS -m abe
1022	#PBS -o $remote_dayfile
1023
1024	$init_cmds
1025	$module_calls
1026
1027	%%END%%
1028
1029	fi
1030
1031	elif [[ $remote_host = lckyut ]]
1032	then
1033	cat > $job_to_send << %%END%%
1034	#!/bin/bash
1035	#PJM -L "vnode=$numprocs"
1036	#PJM -L "vnode-core=1"
1037	#PJM -L "rscgrp=$queue"
1038	#PJM --no-stging
1039	#PJM -L "elapse=$timestring"
1040	#PJM --mpi proc=$numprocs
1041	#PJM -o $remote_dayfile
1042	#PJM -j
1043
1044	export LANG=en_US.UTF-8
1045	%%END%%
1046
1047	elif [[ $remote_host = nech ]]
1048	then
1049
1050	if (( nodes > 1 ))
1051	then
1052	cat > $job_to_send << %%END%%
1053	#!/bin/ksh
1054	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1055	#PBS -l ${qsubmem}=${Memory}gb
1056	#PBS -b $nodes
1057	#PBS -o $remote_dayfile
1058	#PBS -N palm
1059	#PBS -j o
1060	#PBS -T mpisx
1061
1062	%%END%%
1063
1064	elif [[ $numprocs != 0 ]]
1065	then
1066	cat > $job_to_send << %%END%%
1067	#!/bin/ksh
1068	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1069	#PBS -l ${qsubmem}=${Memory}gb
1070	#PBS -o $remote_dayfile
1071	#PBS -N palm
1072	#PBS -j o
1073
1074	%%END%%
1075
1076	else
1077	cat > $job_to_send << %%END%%
1078	#!/bin/ksh
1079	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1080	#PBS -o $remote_dayfile
1081	#PBS -j o
1082
1083	%%END%%
1084
1085	fi
1086
1087	elif [[ $remote_host = necriam ]]
1088	then
1089
1090	if (( nodes > 1 ))
1091	then
1092	cat > $job_to_send << %%END%%
1093	#!/bin/ksh
1094	#PBS -b $nodes
1095	#PBS -o $remote_dayfile
1096	#PBS -N $job_name
1097	#PBS -j o
1098	#PBS -v MPIPROGINV=YES
1099
1100	%%END%%
1101
1102	elif [[ $numprocs != 0 ]]
1103	then
1104	cat > $job_to_send << %%END%%
1105	#!/bin/ksh
1106	#PBS -q ${queue}
1107	#PBS -o $remote_dayfile
1108	#PBS -N $job_name
1109	#PBS -j o
1110	#PBS -v MPIPROGINV=YES
1111
1112	%%END%%
1113
1114	else
1115	cat > $job_to_send << %%END%%
1116	#!/bin/ksh
1117	#PBS -o $remote_dayfile
1118	#PBS -j o
1119
1120	%%END%%
1121
1122	fi
1123
1124	elif [[ $remote_host = lctit ]]
1125	then
1126	cat > $job_to_send << %%END%%
1127	#!/bin/ksh
1128	$init_cmds
1129	$module_calls
1130
1131	%%END%%
1132
1133	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1134	if [[ $tasks_per_node != $processes_per_node ]]
1135	then
1136	submcom="$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 $queue "
1137	else
1138	submcom="$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 "
1139	fi
1140
1141	else
1142
1143	cat > $job_to_send << %%END%%
1144	# @\$-q ${queue}
1145	# @\$-l${qsubtime} $timestring
1146	# @\$-l${qsubmem} ${memory}mb
1147	# @\$-o $remote_dayfile
1148	# @\$-eo
1149
1150	%%END%%
1151
1152	fi
1153
1154
1155	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1156	# BEI EXIT VERANLASSEN
1157	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1158	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1159	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1160	then
1161	echo "set +vx" >> $job_to_send
1162	echo "trap '" >> $job_to_send
1163	echo "set +vx" >> $job_to_send
1164	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 ]]
1165	then
1166	if [[ $remote_host = ibmh ]]
1167	then
1168	return_queue=c1
1169	elif [[ $remote_host = ibmkisti ]]
1170	then
1171	return_queue=class.1-2
1172	elif [[ $remote_host = ibmku ]]
1173	then
1174	return_queue=sdbg2
1175	elif [[ $remote_host = ibms ]]
1176	then
1177	return_queue=p_normal
1178	elif [[ $remote_host = ibmy ]]
1179	then
1180	return_queue=serial
1181	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1182	then
1183	return_queue=serialq
1184	elif [[ $remote_host = necriam ]]
1185	then
1186	return_queue=SP
1187	elif [[ $remote_host = lckiaps ]]
1188	then
1189	return_queue=express
1190	elif [[ $remote_host = lckyut ]]
1191	then
1192	return_queue=cx-single
1193	else
1194	return_queue=unknown
1195	fi
1196
1197	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1198	then
1199
1200	if [[ $remote_host = ibmku ]]
1201	then
1202	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1203	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1204	else
1205	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1206	fi
1207	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1208	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1209	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1210	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1211	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1212	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1213	if [[ $host != "ibmh" ]]
1214	then
1215	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1216	fi
1217	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1218	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1219
1220	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1221	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1222
1223	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1224	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
1225	if [[ $remote_host = ibmku ]]
1226	then
1227	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1228	fi
1229	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1230
1231	elif [[ $remote_host = nech ]]
1232	then
1233	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1234	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1235	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1236	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1237	echo "#PBS -j o" >> $job_to_send
1238	echo " " >> $job_to_send
1239	echo "set -x" >> $job_to_send
1240	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1241	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1242	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1243	echo "%%END%%" >> $job_to_send
1244
1245	elif [[ $remote_host = necriam ]]
1246	then
1247	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1248	echo "#PBS -q $return_queue" >> $job_to_send
1249	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1250	echo "#PBS -j o" >> $job_to_send
1251	echo " " >> $job_to_send
1252	echo "set -x" >> $job_to_send
1253	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1254	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1255	echo "%%END%%" >> $job_to_send
1256
1257	elif [[ $remote_host = lckyut ]]
1258	then
1259	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1260	echo "#!/bin/bash" >> $job_to_send
1261	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1262	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1263	echo "#PJM --no-stging" >> $job_to_send
1264	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1265	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1266	echo "#PJM -j" >> $job_to_send
1267	echo " " >> $job_to_send
1268	echo "export LANG=en_US.UTF-8" >> $job_to_send
1269	echo "set -x" >> $job_to_send
1270	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1271	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1272	echo "%%END%%" >> $job_to_send
1273
1274	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1275	then
1276	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1277	echo "#!/bin/bash" >> $job_to_send
1278	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1279	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1280	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1281	echo "#PBS -l feature=data" >> $job_to_send
1282	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1283	echo "#PBS -j oe" >> $job_to_send
1284	echo " " >> $job_to_send
1285	echo ". /usr/share/modules/init/bash" >> $job_to_send
1286	echo "set -x" >> $job_to_send
1287	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1288	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1289	echo "%%END%%" >> $job_to_send
1290
1291	else
1292
1293	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1294	echo "# @\\\$-q $return_queue" >> $job_to_send
1295	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1296	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1297	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1298	then
1299	echo "# @\$-l mpp_p=0" >> $job_to_send
1300	fi
1301	echo '# @\$-lF 10mb' >> $job_to_send
1302	# echo '# @\$-o /dev/null' >> $job_to_send
1303	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1304	echo '# @\\\$-eo' >> $job_to_send
1305	echo " " >> $job_to_send
1306	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1307	then
1308	echo "set +vx" >> $job_to_send
1309	echo ". .profile" >> $job_to_send
1310	fi
1311	echo "set -x" >> $job_to_send
1312	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
1313	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1314	echo "%%END%%" >> $job_to_send
1315	fi
1316
1317	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1318	then
1319	echo "llsubmit scpjob.$kennung" >> $job_to_send
1320	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1321	then
1322	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1323	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1324	echo "msub scpjob.$kennung" >> $job_to_send
1325	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1326	then
1327	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1328	elif [[ $remote_host = t3es ]]
1329	then
1330	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1331	elif [[ $remote_host = lckiaps ]]
1332	then
1333	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1334	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1335	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1336	elif [[ $remote_host = lckyut ]]
1337	then
1338	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1339	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1340	else
1341	echo "$submcom scpjob.$kennung" >> $job_to_send
1342	fi
1343	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1344	then
1345	echo "rm scpjob.$kennung" >> $job_to_send
1346	fi
1347	if [[ $remote_host = nech ]]
1348	then
1349	echo "cd -" >> $job_to_send
1350	fi
1351	else
1352	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1353	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1354	fi
1355	echo "set -x" >> $job_to_send
1356	echo " ' exit" >> $job_to_send
1357	echo "set -x" >> $job_to_send
1358	fi
1359
1360
1361
1362	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1363	cat $file_to_send >> $job_to_send
1364
1365	if [[ $remote_host = ibm ]]
1366	then
1367	echo " " >> $job_to_send
1368	echo "exit" >> $job_to_send
1369	fi
1370
1371	# remove job file
1372	if [[ $remote_host = lctit \|\| $remote_host = ibmku ]]
1373	then
1374	echo " " >> $job_to_send
1375	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1376	fi
1377
1378
1379
1380	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1381	if [[ -z $remote_user ]]
1382	then
1383	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1384	then
1385	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1386	fi
1387	fi
1388
1389
1390
1391	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1392	if [[ $no_submit = false ]]
1393	then
1394	if [[ $remote_host != $local_host ]]
1395	then
1396	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1397	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1398	then
1399	job_catalog_save=$job_catalog
1400	job_catalog=job_queue
1401	elif [[ $remote_host = nech ]]
1402	then
1403	job_catalog_save=$job_catalog
1404	job_catalog=/hpf/b/${remote_user}/job_queue
1405	fi
1406	if [[ $remote_host = nech ]]
1407	then
1408	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1409	# TRANSFERIERT WERDEN
1410	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1411	else
1412	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1413	fi
1414	if [[ $? = 1 ]]
1415	then
1416	locat=scp; exit
1417	fi
1418	if [[ $remote_host = ibms ]]
1419	then
1420	job_catalog=$job_catalog_save
1421	fi
1422	[[ $verify = true ]] && printf "\n >>> finished\n"
1423	else
1424	eval job_catalog=$job_catalog
1425	cp $job_to_send ${job_catalog}/$job_on_remhost
1426	fi
1427
1428
1429
1430	# NQS- BZW. LOADLEVELER-JOB STARTEN
1431	if [[ $remote_host != $local_host ]]
1432	then
1433	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1434
1435	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1436	then
1437	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1438	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1439	elif [[ $remote_host = ibmku ]]
1440	then
1441	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1442	else
1443	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1444	fi
1445
1446	[[ $verify = true ]] && printf " >>> o.k.\n"
1447	else
1448	cd $job_catalog
1449	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1450	then
1451	eval $submcom $job_on_remhost
1452	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyut ]]
1453	then
1454	chmod u+x $job_on_remhost
1455	echo "$submcom $job_on_remhost"
1456	eval $submcom $job_on_remhost
1457	elif [[ $local_host = nech ]]
1458	then
1459	if [[ $queue = default ]]
1460	then
1461	eval $submcom $job_on_remhost
1462	else
1463	eval $submcom -q $queue $job_on_remhost
1464	fi
1465	else
1466	qsub $job_on_remhost
1467	fi
1468
1469	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1470	# only at the end of the job.
1471	if [[ $local_host != lctit && $local_host != ibmku ]]
1472	then
1473	rm $job_on_remhost
1474	fi
1475	cd - > /dev/null
1476	fi
1477	fi
1478
1479
1480
1481	# ABSCHLUSSARBEITEN
1482
1483	if [[ $no_submit = false ]]
1484	then
1485	rm $job_to_send
1486	fi
1487	[[ $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 |