Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1260

Last change on this file since 1260 was 1260, checked in by raasch, 11 years ago
host jaboticaba admitted for subjob
Property svn:keywords set to `Id Rev`
File size: 58.2 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	# jaboticaba admitted
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1260 2013-12-04 12:48:04Z raasch $
30	#
31	# 1255 2013-11-07 14:43:35Z raasch
32	# further adjustments for lccrayb
33	#
34	# 1224 2013-09-16 07:27:23Z raasch
35	# first adjustments for lccrayb
36	#
37	# 1202 2013-07-10 16:22:07Z witha
38	# adjustments for Forwind cluster (lcflow)
39	#
40	# 1199 2013-07-05 14:52:22Z raasch
41	# adjustments for CSC Helsinki (lccrayf)
42	#
43	# use of cluster/express queue enabled (ibmh)
44	# vinessa added (imuk)
45	#
46	# 1103 2013-02-20 02:15:53Z raasch
47	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
48	# instead typeset -L/R),
49	# further adjustments for lckyuh
50	#
51	# 2013-02-10 01:47:43Z raasch
52	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
53	# and for Forwind cluster (lcflow)
54	#
55	# 1094 2013-02-03 01:52:12Z raasch
56	# new option -P for explicit setting of ssh/scp port,
57	# decalpha parts (yonsei) removed
58	#
59	# 2013-02-02 07:06:13Z raasch
60	# adjustments for Kyushu-University computing center (lckyut - tatara)
61	# old changelog messages removed
62	#
63	# 1046 2012-11-09 14:38:45Z maronga
64	# code put under GPL (PALM 3.9)
65	#
66	# 08/07/94 - Siggi - first version finished
67	# 29/06/94 - Siggi - script development started
68	#--------------------------------------------------------------------------------#
69	# subjob - script for automatic generation and submission of batch-job files
70	# for various batch queuing systems
71	#--------------------------------------------------------------------------------#
72
73
74	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
75	delete_dayfile=false
76	email_notification=none
77	group_number=none
78	locat=normal
79	no_default_queue=none
80	no_submit=false
81	job_catalog="~/job_queue"
82	job_name=none
83	local_user=$LOGNAME
84	node_usage=shared
85	numprocs=0
86	punkte="..........................................................."
87	submcom=qsub
88	queue=default
89	remote_host=none
90	remote_user=""
91	verify=true
92
93	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
94	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
95
96
97
98	# FEHLERBEHANDLUNG
99	# BEI EXIT:
100	trap 'if [[ $locat != normal ]]
101	then
102	case $locat in
103	(option) printf "\n --> available optios can be displayed"
104	printf " by typing:"
105	printf "\n \"subjob ?\" \n";;
106	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
107	(*) printf "\n +++ unknown error"
108	printf "\n please inform S. Raasch!\n"
109	esac
110	[[ -f $job_to_send ]] && rm $job_to_send
111	printf "\n\n+++ SUBJOB killed \n\n"
112	fi' exit
113
114
115	# BEI TERMINAL-BREAK:
116	trap '[[ -f $job_to_send ]] && rm $job_to_send
117	printf "\n\n+++ SUBJOB killed \n\n"
118	exit
119	' 2
120
121
122
123
124	# LOKALEN HOSTNAMEN FESTSTELLEN
125	local_host=$(hostname)
126
127
128
129	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
130	# UEBERHAUPT ZULAESSIG IST
131	# Note: One of the entries for "lck" or "lckordi" always has to be
132	# comment out, because the hostname (node*) is same for both machines
133	case $local_host in
134	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
135	(atmos) local_addres=172.20.25.35; local_host=lcide;;
136	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
137	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
138	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
139	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
140	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
141	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
142	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
143	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
144	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
145	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
146	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
147	(blogin\|bxc) local_addres=130.73.233.1; local_host=lccrayb;;
148	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
149	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
150	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
151	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
152	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
153	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
154	(cs*) local_addres=136.172.44.131; local_host=nech;;
155	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
156	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
157	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
158	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
159	(node*) local_addres=165.132.26.61 local_host=lck;;
160	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
161	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
162	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
163	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
164	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
165	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
166	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
167	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
168	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
169	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
170	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
171	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
172	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
173	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
174	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
175	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
176	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
177	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
178	(jaboticaba) local_addres=150.163.25.181; local_host=lcbr;;
179	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
180	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
181	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
182	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
183	(meller) local_addres=134.106.74.155; local_host=lcfor;;
184	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
185	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
186	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
187	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
188	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
189	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
190	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
191	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
192	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
193	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
194	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
195	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
196	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
197	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
198	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
199	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
200	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
201	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
202	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
203	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
204	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
205	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
206	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
207	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
208	(*) printf "\n +++ \"$local_host\" unknown";
209	printf "\n please contact the PALM group at IMUK";
210	locat=parameter; exit;;
211	esac
212
213
214
215	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
216	remote_host=$local_host
217
218
219
220
221	# PROZEDUROPTIONEN EINLESEN
222	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
223	do
224	case $option in
225	(c) job_catalog=$OPTARG;;
226	(d) delete_dayfile=true;;
227	(D) no_submit=true;;
228	(e) email_notification=$OPTARG;;
229	(g) group_number=$OPTARG;;
230	(h) remote_host=$OPTARG;;
231	(m) memory=$OPTARG;;
232	(n) job_name=$OPTARG;;
233	(N) node_usage=$OPTARG;;
234	(O) threads_per_task=$OPTARG;;
235	(P) scp_port=$OPTARG;;
236	(q) no_default_queue=$OPTARG;;
237	(t) cputime=$OPTARG;;
238	(T) tasks_per_node=$OPTARG;;
239	(u) remote_user=$OPTARG;;
240	(v) verify=false;;
241	(X) numprocs=$OPTARG;;
242	(\?) printf "\n +++ Option $OPTARG unknown \n";
243	locat=option; exit;;
244	esac
245	done
246
247
248	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
249	(( to_shift = $OPTIND - 1 ))
250	shift $to_shift; file_to_send=$1
251
252
253	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
254	if [ "$1" = "?" ]
255	then
256	(printf "\n *** subjob can be called as follows:\n"
257	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
258	printf "\n Description of available options:\n"
259	printf "\n Option Description Default-Value"
260	printf "\n -c job-input- and output-catalog ~/job_queue"
261	printf "\n -d no job-protocol will be created ---"
262	printf "\n -D only the job-file will be created ---"
263	printf "\n -h execution host, available hosts: $remote_host"
264	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
265	printf "\n lckiaps, lctit, nech, necriam, unics"
266	printf "\n -m memory demand per process in MByte ---"
267	printf "\n -n jobname <jobdatei>"
268	printf "\n -O threads per task (for OpenMP usage) 1"
269	printf "\n -P ssh/scp port default port"
270	printf "\n -q job-queue to be used default"
271	printf "\n -t allowed cpu-time in seconds ---"
272	printf "\n -T tasks per node (on parallel hosts) ---"
273	printf "\n -u username on execution host from .netrc"
274	printf "\n -v no prompt for confirmation ---"
275	printf "\n -X # of processors (on parallel hosts) 1"
276	printf "\n "
277	printf "\n The only possible positional parameter is <jobfile>:"
278	printf "\n The complete NQS-job must be provided here."
279	printf "\n <jobfile>=? creates this outline\n\n") \| more
280	exit
281	fi
282
283
284
285	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
286	if [[ "$file_to_send" = "" ]]
287	then
288	printf "\n +++ job-file missing"
289	locat=parameter; exit
290	else
291	if [[ -f $file_to_send ]]
292	then
293	true
294	else
295	printf "\n +++ job-file: "
296	printf "\n $file_to_send"
297	printf "\n does not exist"
298	locat=parameter; exit
299	fi
300	fi
301
302
303
304	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
305	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
306	if [[ $job_name = none ]]
307	then
308	job_name=$file_to_send
309	fi
310	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
311	then
312	printf "\n +++ job-file name: "
313	printf "\n $job_name"
314	printf "\n must not contain \"/\"-characters"
315	locat=parameter; exit
316	fi
317
318
319
320
321	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
322	# ODER WENN HOST NICHT ANGEGEBEN WURDE
323	if [[ $remote_host = none ]]
324	then
325	printf "\n +++ host missing"
326	locat=option; exit
327	else
328	case $remote_host in
329	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
330	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
331	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
332	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
333	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
334	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
335	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/torque/4.2.4.1/bin/qsub;;
336	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
337	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
338	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
339	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
340	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
341	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
342	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
343	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
344	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
345	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
346	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
347	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
348	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
349	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
350	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
351	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
352	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
353	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
354	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
355	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
356	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
357	locat=parameter; exit;;
358	esac
359	fi
360
361
362	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
363	if [[ $no_default_queue != none ]]
364	then
365	error=false
366	ndq=$no_default_queue
367	case $remote_host in
368	(ibm) case $ndq in
369	(p690_express\|p690_standard\|p690_long) error=false;;
370	(*) error=true;;
371	esac;;
372	(ibmh) case $ndq in
373	(cluster\|express) error=false;;
374	(*) error=true;;
375	esac;;
376	(ibmkisti) case $ndq in
377	(class.32plus\|class.1-2\|class.2-32) error=false;;
378	(*) error=true;;
379	esac;;
380	(ibmku) case $ndq in
381	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
382	(*) error=true;;
383	esac;;
384	(ibms) case $ndq in
385	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
386	(*) error=true;;
387	esac;;
388	(ibmy) case $ndq in
389	(parallel) error=false;;
390	(*) error=true;;
391	esac;;
392	(lccrayb) case $ndq in
393	(mpp1q\|mpp1testq\|testq) error=false;;
394	(*) error=true;;
395	esac;;
396	(lccrayf) case $ndq in
397	(usup\|test*\|small\|large) error=false;;
398	(*) error=true;;
399	esac;;
400	(lcflow) case $ndq in
401	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
402	(*) error=true;;
403	esac;;
404	(lckiaps) case $ndq in
405	(express\|normal) error=false;;
406	(*) error=true;;
407	esac;;
408	(lckyoto) case $ndq in
409	(eh\|ph) error=false;;
410	(*) error=true;;
411	esac;;
412	(lckyuh) case $ndq in
413	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
414	(*) error=true;;
415	esac;;
416	(lckyut) case $ndq in
417	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
418	(*) error=true;;
419	esac;;
420	(lcsgib\|lcsgih) case $ndq in
421	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
422	(*) error=true;;
423	esac;;
424	(lctit) case $ndq in
425	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
426	(*) error=true;;
427	esac;;
428	(t3eb) case $ndq in
429	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
430	(*) error=true;;
431	esac;;
432	(necriam) case $ndq in
433	(SP\|SS\|P6) error=false;;
434	(*) error=true;;
435	esac;;
436	(t3eh) case $ndq in
437	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
438	(*) error=true;;
439	esac;;
440	(t3ej2\|t3ej5) case $ndq in
441	(low\|normal\|high) error=false;;
442	(*) error=true;;
443	esac;;
444	(t3es) case $ndq in
445	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
446	(*) error=true;;
447	esac;;
448	(unics) case $ndq in
449	(unics\|ht) error=false;;
450	(*) error=true;;
451	esac;;
452	esac
453	if [[ $error = true ]]
454	then
455	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
456	locat=parameter; exit
457	else
458	queue=$no_default_queue
459	fi
460	fi
461
462
463
464	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
465	# AUFTEILEN
466	done=false
467	while [[ $done = false ]]
468	do
469	if (( $cputime <= 0 ))
470	then
471	printf "\n +++ wrong cpu-time or cpu-time missing"
472	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
473	printf "\n >>> "
474	read cputime 1>/dev/null 2>&1
475	else
476	done=true
477	fi
478	done
479	if [[ $remote_host = nech ]]
480	then
481	if (( tasks_per_node != 0 ))
482	then
483	(( cputime = cputime * tasks_per_node ))
484	elif [[ $numprocs != 0 ]]
485	then
486	(( cputime = cputime * numprocs ))
487	fi
488	fi
489	(( stunden = cputime / 3600 ))
490	(( resttime = cputime - stunden * 3600 ))
491	(( minuten = resttime / 60 ))
492	(( sekunden = resttime - minuten * 60 ))
493	timestring=${stunden}:${minuten}:${sekunden}
494
495
496
497	# PRUEFEN DER KERNSPEICHERANFORDERUNG
498	done=false
499	while [[ $done = false ]]
500	do
501	if (( memory <= 0 ))
502	then
503	printf "\n +++ wrong memory demand or memory demand missing"
504	printf "\n >>> Please type memory in MByte per process as INTEGER:"
505	printf "\n >>> "
506	read memory 1>/dev/null 2>&1
507	else
508	done=true
509	fi
510	done
511
512	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
513	then
514	if (( tasks_per_node != 0 ))
515	then
516	(( Memory = memory * tasks_per_node / 1000 ))
517	elif [[ $numprocs != 0 ]]
518	then
519	(( Memory = memory * numprocs / 1000 ))
520	else
521	(( Memory = memory / 1000 ))
522	fi
523	elif [[ $remote_host = lctit ]]
524	then
525	(( Memory = memory * tasks_per_node / 1000 ))
526	fi
527
528
529	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
530	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
531	then
532	(( memory = memory * threads_per_task ))
533	fi
534
535
536	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
537	if (( tasks_per_node != 0 ))
538	then
539	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
540	fi
541
542
543	# Calculate number of processes per node
544	(( processes_per_node = tasks_per_node * threads_per_task ))
545
546
547	# Calculate number of MPI tasks
548	(( mpi_tasks = numprocs / threads_per_task ))
549
550
551	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
552	if [[ "$scp_port" != "" ]]
553	then
554	PORTOPT="-P $scp_port"
555	SSH_PORTOPT="-p $scp_port"
556	fi
557
558
559	# HEADER-AUSGABE
560	if [[ $verify = true ]]
561	then
562	printf "\n\n"
563	printf "#--------------------------------------------------------------# \n"
564	spalte1=SUBJOB;spalte2=$(date)
565	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
566	printf "\| \| \n"
567	printf "\| values of parameters/options: \| \n"
568	spalte1=$(echo local_host$punkte \| cut -c-20)
569	spalte2=$punkte$local_host
570	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
571	spalte1=$(echo remote_host$punkte \| cut -c-20)
572	spalte2=$punkte$remote_host
573	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
574	spalte1=$(echo queue$punkte \| cut -c-20)
575	spalte2=$punkte$queue
576	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
577	spalte1=$(echo memory$punkte \| cut -c-20)
578	spalte2="$punkte$memory mb"
579	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
580	spalte1=$(echo cputime$punkte \| cut -c-20)
581	spalte2="$punkte$cputime sec"
582	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
583	spalte1=$(echo job_name$punkte \| cut -c-20)
584	spalte2="$punkte$job_name"
585	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
586	printf "#--------------------------------------------------------------# \n\n"
587
588
589	# KONTROLLABFRAGE, OB ALLES O.K.
590	antwort="dummy"
591	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
592	do
593	read antwort?" >>> continue (y/n) ? "
594	done
595	if [[ $antwort = n \|\| $antwort = N ]]
596	then
597	locat=verify; exit
598	fi
599	printf "\n"
600	fi
601
602
603
604	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
605	kennung=$RANDOM
606	job_on_remhost=${job_name}_${kennung}_$local_host
607	job_to_send=job_to_send_$kennung
608	if [[ $delete_dayfile = false ]]
609	then
610	remote_dayfile=${local_host}_${job_name}_result_$kennung
611	local_dayfile=${remote_host}_${job_name}
612	else
613	remote_dayfile=/dev/null
614	fi
615
616
617
618	# Generate the batch job scripts (qsub/msub/LoadLeveler)
619	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
620	then
621
622	# General LoadLeveler settings
623	execute_in_shell="#!/bin/ksh"
624	use_shell="# @ shell = /bin/ksh"
625	consumable_memory="ConsumableMemory($memory mb)"
626	class="# @ class = $queue"
627	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
628	network_to_use="# @ network.mpi = sn_all,shared,us"
629	data_limit="# @ data_limit = 1.76gb"
630	image_size="# @ image_size = 50"
631	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
632
633	if [[ $email_notification = none ]]
634	then
635	notify_user=""
636	else
637	notify_user="# @ notify_user = $email_notification"
638	if [[ $delete_dayfile = true ]]
639	then
640	notification='# @ notification = never'
641	fi
642	fi
643
644	if [[ $remote_host = ibmh ]]
645	then
646	data_limit=""
647	network_to_use=""
648	class="# @ class = $queue"
649	environment=""
650	rset="# @ rset = RSET_MCM_AFFINITY"
651	task_affinity="# @ task_affinity = core(1)"
652	elif [[ $remote_host = ibmkisti ]]
653	then
654	network_to_use="# @ network.MPI = sn_all,shared,US"
655	wall_clock_limit="# @ wall_clock_limit = $timestring"
656	if [[ $threads_per_task = 1 ]]
657	then
658	rset="# @ rset = RSET_MCM_AFFINITY"
659	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
660	fi
661	environment=""
662	use_shell=""
663	data_limit=""
664	image_size=""
665	elif [[ $remote_host = ibmku ]]
666	then
667	execute_in_shell="#!/usr/bin/ksh"
668	use_shell="# @ shell = /usr/bin/ksh"
669	consumable_memory=""
670	environment=""
671	network_to_use="# @ network.mpi = sn_all,shared,us"
672	data_limit=""
673	image_size=""
674	elif [[ $remote_host = ibms ]]
675	then
676	network_to_use="# @ network.mpi = csss,shared,us"
677	elif [[ $remote_host = ibmy ]]
678	then
679	consumable_memory=""
680	network_to_use=""
681	fi
682
683	cat > $job_to_send << %%END%%
684	$execute_in_shell
685	$use_shell
686
687	# @ job_type = parallel
688	# @ job_name = $job_name
689	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
690	# @ output = $remote_dayfile
691	# @ error = $remote_dayfile
692	$wall_clock_limit
693	$image_size
694	$class
695	$environment
696	$network_to_use
697	$data_limit
698	$rset
699	$mcm_affinity_options
700	$task_affinity
701	$notification
702	$notify_user
703
704	%%END%%
705
706	if (( nodes > 0 ))
707	then
708
709	if [[ $remote_host != ibmkisti ]]
710	then
711
712	cat >> $job_to_send << %%END%%
713	# @ node = $nodes
714	# @ tasks_per_node = $processes_per_node
715	# @ node_usage = $node_usage
716	# @ queue
717
718	%%END%%
719
720	else
721
722	cat >> $job_to_send << %%END%%
723	# @ total_tasks = $mpi_tasks
724	# @ blocking = unlimited
725	# @ queue
726
727	%%END%%
728
729	fi
730
731	else
732
733	if [[ $remote_host != ibmy ]]
734	then
735
736	cat >> $job_to_send << %%END%%
737	# @ blocking = unlimited
738	# @ total_tasks = $numprocs
739	# @ node_usage = $node_usage
740	# @ queue
741
742	%%END%%
743
744	else
745
746	cat >> $job_to_send << %%END%%
747	# @ node = 1
748	# @ total_tasks = $numprocs
749	# @ queue
750
751	%%END%%
752
753	fi
754
755	fi
756
757	# workaround because of silly job filter on ibmkisti
758	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
759	then
760	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
761	fi
762
763	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
764	then
765
766	cat > $job_to_send << %%END%%
767	#!/bin/ksh
768
769	# @ job_type = serial
770	# @ node_usage = $node_usage
771	# @ job_name = palm
772	# @ wall_clock_limit = ${timestring},$timestring
773	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
774	# @ output = $remote_dayfile
775	# @ error = $remote_dayfile
776	$class
777	$notification
778
779	# @ queue
780
781	%%END%%
782
783	elif [[ $remote_host = lccrayb ]]
784	then
785
786	if [[ $numprocs != 0 ]]
787	then
788	cat > $job_to_send << %%END%%
789	#!/bin/bash
790	#PBS -N $job_name
791	#PBS -l walltime=$timestring
792	#PBS -l mppwidth=$numprocs
793	#PBS -l mppnppn=$processes_per_node
794	#PBS -o $remote_dayfile
795	#PBS -j oe
796	#PBS -q $queue
797
798	$module_calls
799
800	%%END%%
801
802	else
803
804	continue
805
806	fi
807
808	elif [[ $remote_host = lccrayf ]]
809	then
810
811	if [[ $numprocs != 0 ]]
812	then
813	cat > $job_to_send << %%END%%
814	#!/bin/bash -l
815	#SBATCH -J $job_name
816	#SBATCH -t $timestring
817	#SBATCH -N $nodes
818	#SBATCH --ntasks-per-node=$processes_per_node
819	#SBATCH -p $queue
820	#SBATCH -o $remote_dayfile
821	#SBATCH -e $remote_dayfile
822
823	$init_cmds
824	$module_calls
825
826	%%END%%
827
828	else
829	cat > $job_to_send << %%END%%
830	#!/bin/bash -l
831	#SBATCH -J $job_name
832	#SBATCH -t $timestring
833	#SBATCH -l ncpus=1
834	#SBATCH -l pmem=${memory}mb
835	#SBATCH -m abe
836	#SBATCH -o $remote_dayfile
837	#SBATCH -e $remote_dayfile
838
839	$init_cmds
840	$module_calls
841
842	%%END%%
843
844	fi
845
846	elif [[ $remote_host = lcfimm ]]
847	then
848
849	if [[ $numprocs != 0 ]]
850	then
851	cat > $job_to_send << %%END%%
852	#!/bin/ksh
853	#PBS -N $job_name
854	#PBS -A $project_account
855	#PBS -l walltime=$timestring
856	#PBS -l nodes=${nodes}:ppn=$processes_per_node
857	#PBS -l pmem=${memory}mb
858	#PBS -m abe
859	#PBS -o $remote_dayfile
860	#PBS -j oe
861	mpd &
862
863	%%END%%
864
865	else
866	cat > $job_to_send << %%END%%
867	#!/bin/ksh
868	#PBS -N $job_name
869	#PBS -A $project_account
870	#PBS -l walltime=$timestring
871	#PBS -l ncpus=1
872	#PBS -l pmem=${memory}mb
873	#PBS -m abe
874	#PBS -o $remote_dayfile
875	#PBS -j oe
876
877	%%END%%
878
879	fi
880
881	elif [[ $remote_host = lcflow ]]
882	then
883	if [ $memory -gt 1800 ]; then
884	use_himem=""
885	else
886	use_himem="#"
887	fi
888	if [[ $numprocs != 0 ]]
889	then
890	pe_set="#$ -pe impi41 $numprocs"
891	else
892	pe_set="#$ -pe impi41 1"
893	fi
894	if [[ $queue = default ]]
895	then
896	queue_set=""
897	else
898	queue_set="#$ -q $queue"
899	fi
900	[[ "$disc_space" = "" ]] && disc_space=50
901
902	cat > $job_to_send << %%END%%
903	#!/bin/bash
904	#$ -S /bin/bash
905	#$ -N $job_name
906	#$ -cwd
907	#$ -l h_rt=$timestring
908	#$ -l h_vmem=${memory}M
909	#$ -o $remote_dayfile
910	#$ -j y
911	$pe_set
912	#$ -R y
913	${use_himem}#$ -l highmem=true
914	#$ -l h_fsize=${disc_space}G
915	$queue_set
916
917	%%END%%
918
919	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
920	then
921
922	if [[ $numprocs != 0 ]]
923	then
924	cat > $job_to_send << %%END%%
925	#!/bin/ksh
926	#PBS -N $job_name
927	#PBS -l walltime=$timestring
928	#PBS -l ncpus=$numprocs
929	#PBS -l pmem=${memory}mb
930	#PBS -o $remote_dayfile
931	#PBS -l nodes=$nodes:ppn=${processes_per_node}
932	#PBS -j oe
933
934	mpd &
935
936	%%END%%
937
938	else
939	cat > $job_to_send << %%END%%
940	#!/bin/ksh
941	#PBS -N $job_name
942	#PBS -l walltime=$timestring
943	#PBS -l ncpus=1
944	#PBS -l pmem=${memory}mb
945	#PBS -o $remote_dayfile
946	#PBS -j oe
947
948	%%END%%
949
950	fi
951
952	elif [[ $remote_host = lckiaps ]]
953	then
954
955	if [[ $numprocs != 0 ]]
956	then
957	cat > $job_to_send << %%END%%
958	#!/bin/ksh
959	#PBS -N $job_name
960	#PBS -l walltime=$timestring
961	#PBS -l select=1:ncpus=$numprocs
962	#PBS -l pmem=${memory}mb
963	#PBS -q $queue
964	#PBS -o $remote_dayfile
965	#PBS -j oe
966	#PBS -V
967
968	%%END%%
969
970	else
971	cat > $job_to_send << %%END%%
972	#!/bin/ksh
973	#PBS -N $job_name
974	#PBS -l walltime=$timestring
975	#PBS -l ncpus=1
976	#PBS -l pmem=${memory}mb
977	#PBS -o $remote_dayfile
978	#PBS -j oe
979
980	%%END%%
981
982	fi
983
984	elif [[ $remote_host = lcyon ]]
985	then
986
987	if [[ $numprocs != 0 ]]
988	then
989	cat > $job_to_send << %%END%%
990	#!/bin/ksh
991	#PBS -N $job_name
992	#PBS -l walltime=$timestring
993	#PBS -l ncpus=$numprocs
994	#PBS -l pmem=${memory}mb
995	#PBS -o $remote_dayfile
996	#PBS -j oe
997
998	%%END%%
999
1000	else
1001	cat > $job_to_send << %%END%%
1002	#!/bin/ksh
1003	#PBS -N $job_name
1004	#PBS -l walltime=$timestring
1005	#PBS -l ncpus=1
1006	#PBS -l pmem=${memory}mb
1007	#PBS -o $remote_dayfile
1008	#PBS -j oe
1009
1010	%%END%%
1011
1012	fi
1013
1014	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1015	then
1016
1017	if [[ "$sgi_feature" != "" ]]
1018	then
1019	feature_directive="#PBS -l feature=$sgi_feature"
1020	else
1021	feature_directive=""
1022	fi
1023
1024	if [[ $queue = dataq \|\| $queue = permq ]]
1025	then
1026	feature_directive="#PBS -l feature=data"
1027	fi
1028
1029	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
1030	then
1031	queue_directive="#PBS -q $queue"
1032	else
1033	queue_directive=""
1034	fi
1035
1036	if [[ $email_notification = none ]]
1037	then
1038	email_directive=""
1039	else
1040	email_directive="#PBS -M $email_notification"
1041	fi
1042
1043	if [[ $numprocs != 0 ]]
1044	then
1045	cat > $job_to_send << %%END%%
1046	#!/bin/bash
1047	#PBS -N $job_name
1048	#PBS -l walltime=$timestring
1049	#PBS -l nodes=$nodes:ppn=${processes_per_node}
1050	#PBS -l naccesspolicy=$node_usage
1051	#PBS -o $remote_dayfile
1052	#PBS -j oe
1053	$feature_directive
1054	$queue_directive
1055	$email_directive
1056
1057	eval \`/sw/swdist/bin/modulesinit\`
1058	#. /usr/share/modules/init/bash
1059	$init_cmds
1060	$module_calls
1061
1062	echo ld_library_path=\$LD_LIBRARY_PATH
1063
1064	%%END%%
1065
1066	else
1067	cat > $job_to_send << %%END%%
1068	#PBS -S /bin/bash
1069	#PBS -N $job_name
1070	#PBS -l walltime=$timestring
1071	#PBS -l ncpus=1
1072	#PBS -o $remote_dayfile
1073	#PBS -j oe
1074	$queue_directive
1075	$email_directive
1076
1077	eval \`/sw/swdist/bin/modulesinit\`
1078	#. /usr/share/modules/init/bash
1079	$init_cmds
1080	$module_calls
1081
1082	%%END%%
1083
1084	fi
1085
1086	elif [[ $remote_host = lcxe6 ]]
1087	then
1088
1089	if [[ $numprocs != 0 ]]
1090	then
1091	cat > $job_to_send << %%END%%
1092	#!/bin/ksh
1093	#PBS -S /bin/ksh
1094	#PBS -N $job_name
1095	#PBS -A $project_account
1096	#PBS -j oe
1097	#PBS -l walltime=$timestring
1098	#PBS -l mppwidth=${numprocs}
1099	#PBS -l mppnppn=${processes_per_node}
1100	#PBS -m abe
1101	#PBS -o $remote_dayfile
1102	$email_directive
1103
1104	$init_cmds
1105	$module_calls
1106
1107	%%END%%
1108
1109	else
1110	cat > $job_to_send << %%END%%
1111	#!/bin/ksh
1112	#PBS -S /bin/ksh
1113	#PBS -N $job_name
1114	#PBS -A $project_account
1115	#PBS -j oe
1116	#PBS -l walltime=$timestring
1117	#PBS -l ncpus=1
1118	#PBS -l pmem=${memory}mb
1119	#PBS -m abe
1120	$email_directive
1121	#PBS -o $remote_dayfile
1122
1123	$init_cmds
1124	$module_calls
1125
1126	%%END%%
1127
1128	fi
1129
1130	elif [[ $remote_host = lckyoto ]]
1131	then
1132
1133	cat > $job_to_send << %%END%%
1134	#!/bin/ksh
1135	# @\$-o $remote_dayfile
1136	# @\$-eo -oi
1137	# @\$-lP 16
1138	# @\$-lp 1
1139	# @\$-lm 28gb -llm unlimited -ls unlimited
1140	# @\$-q $queue
1141	# @\$-Pvn abs_pack
1142	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1143	#. /thin/local/etc/setprofile/intel-11.0.sh
1144	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1145	. ~/.myprofile
1146	#. /home2/t/t51254/palm/current_version/myprofile
1147	#. /thin/apps/pgi/mpi.sh
1148	#
1149	env
1150	#
1151	set -x
1152
1153	%%END%%
1154
1155	elif [[ $remote_host = lcxt5m ]]
1156	then
1157
1158	if [[ $numprocs != 0 ]]
1159	then
1160	cat > $job_to_send << %%END%%
1161	#!/bin/ksh
1162	#PBS -S /bin/ksh
1163	#PBS -N $job_name
1164	#PBS -j oe
1165	#PBS -l walltime=$timestring
1166	#PBS -l mppwidth=${numprocs}
1167	#PBS -l mppnppn=${processes_per_node}
1168	#PBS -m abe
1169	#PBS -o $remote_dayfile
1170
1171	$init_cmds
1172	$module_calls
1173
1174	%%END%%
1175
1176	else
1177	cat > $job_to_send << %%END%%
1178	#!/bin/ksh
1179	#PBS -S /bin/ksh
1180	#PBS -N $job_name
1181	#PBS -j oe
1182	#PBS -l walltime=$timestring
1183	#PBS -l ncpus=1
1184	#PBS -l pmem=${memory}mb
1185	#PBS -m abe
1186	#PBS -o $remote_dayfile
1187
1188	$init_cmds
1189	$module_calls
1190
1191	%%END%%
1192
1193	fi
1194
1195	elif [[ $remote_host = lckyuh ]]
1196	then
1197	cat > $job_to_send << %%END%%
1198	#!/bin/bash
1199	#PJM -L "rscgrp=$queue"
1200	#PJM -L "node=$nodes"
1201	#PJM --mpi "proc=$numprocs"
1202	#PJM -L "elapse=$timestring"
1203	#PJM -o $remote_dayfile
1204	#PJM -j
1205	#PJM -X
1206	#PJM --no-stging
1207
1208	export LANG=en_US.UTF-8
1209	%%END%%
1210
1211	elif [[ $remote_host = lckyut ]]
1212	then
1213	cat > $job_to_send << %%END%%
1214	#!/bin/bash
1215	#PJM -L "rscgrp=$queue"
1216	#PJM -L "vnode=$numprocs"
1217	#PJM -L "vnode-core=1"
1218	#PJM -L "elapse=$timestring"
1219	#PJM --mpi proc=$numprocs
1220	#PJM -o $remote_dayfile
1221	#PJM -j
1222	#PJM -X
1223	#PJM --no-stging
1224
1225	export LANG=en_US.UTF-8
1226	%%END%%
1227
1228	elif [[ $remote_host = nech ]]
1229	then
1230
1231	if (( nodes > 1 ))
1232	then
1233	cat > $job_to_send << %%END%%
1234	#!/bin/ksh
1235	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1236	#PBS -l ${qsubmem}=${Memory}gb
1237	#PBS -b $nodes
1238	#PBS -o $remote_dayfile
1239	#PBS -N palm
1240	#PBS -j o
1241	#PBS -T mpisx
1242
1243	%%END%%
1244
1245	elif [[ $numprocs != 0 ]]
1246	then
1247	cat > $job_to_send << %%END%%
1248	#!/bin/ksh
1249	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1250	#PBS -l ${qsubmem}=${Memory}gb
1251	#PBS -o $remote_dayfile
1252	#PBS -N palm
1253	#PBS -j o
1254
1255	%%END%%
1256
1257	else
1258	cat > $job_to_send << %%END%%
1259	#!/bin/ksh
1260	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1261	#PBS -o $remote_dayfile
1262	#PBS -j o
1263
1264	%%END%%
1265
1266	fi
1267
1268	elif [[ $remote_host = necriam ]]
1269	then
1270
1271	if (( nodes > 1 ))
1272	then
1273	cat > $job_to_send << %%END%%
1274	#!/bin/ksh
1275	#PBS -b $nodes
1276	#PBS -o $remote_dayfile
1277	#PBS -N $job_name
1278	#PBS -j o
1279	#PBS -v MPIPROGINV=YES
1280
1281	%%END%%
1282
1283	elif [[ $numprocs != 0 ]]
1284	then
1285	cat > $job_to_send << %%END%%
1286	#!/bin/ksh
1287	#PBS -q ${queue}
1288	#PBS -o $remote_dayfile
1289	#PBS -N $job_name
1290	#PBS -j o
1291	#PBS -v MPIPROGINV=YES
1292
1293	%%END%%
1294
1295	else
1296	cat > $job_to_send << %%END%%
1297	#!/bin/ksh
1298	#PBS -o $remote_dayfile
1299	#PBS -j o
1300
1301	%%END%%
1302
1303	fi
1304
1305	elif [[ $remote_host = lctit ]]
1306	then
1307	cat > $job_to_send << %%END%%
1308	#!/bin/ksh
1309	$init_cmds
1310	$module_calls
1311
1312	%%END%%
1313
1314	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1315	if [[ $tasks_per_node != $processes_per_node ]]
1316	then
1317	submcom="$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 $queue "
1318	else
1319	submcom="$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 "
1320	fi
1321
1322	else
1323
1324	cat > $job_to_send << %%END%%
1325	# @\$-q ${queue}
1326	# @\$-l${qsubtime} $timestring
1327	# @\$-l${qsubmem} ${memory}mb
1328	# @\$-o $remote_dayfile
1329	# @\$-eo
1330
1331	%%END%%
1332
1333	fi
1334
1335
1336	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1337	# BEI EXIT VERANLASSEN
1338	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1339	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1340	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1341	then
1342	echo "set +vx" >> $job_to_send
1343	echo "trap '" >> $job_to_send
1344	echo "set +vx" >> $job_to_send
1345	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $remote_host = lccrayb \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lcflow \|\| $remote_host = lckiaps \|\| $remote_host = lckyu* ]]
1346	then
1347	if [[ $remote_host = ibmh ]]
1348	then
1349	return_queue=c1
1350	elif [[ $remote_host = ibmkisti ]]
1351	then
1352	return_queue=class.1-2
1353	elif [[ $remote_host = ibmku ]]
1354	then
1355	return_queue=sdbg2
1356	elif [[ $remote_host = ibms ]]
1357	then
1358	return_queue=p_normal
1359	elif [[ $remote_host = ibmy ]]
1360	then
1361	return_queue=serial
1362	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1363	then
1364	return_queue=serialq
1365	elif [[ $remote_host = necriam ]]
1366	then
1367	return_queue=SP
1368	elif [[ $remote_host = lccrayb ]]
1369	then
1370	return_queue=dataq
1371	elif [[ $remote_host = lckiaps ]]
1372	then
1373	return_queue=express
1374	elif [[ $remote_host = lckyuh ]]
1375	then
1376	return_queue=cx-single
1377	elif [[ $remote_host = lckyut ]]
1378	then
1379	return_queue=cx-single
1380	else
1381	return_queue=unknown
1382	fi
1383
1384	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1385	then
1386
1387	if [[ $remote_host = ibmku ]]
1388	then
1389	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1390	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1391	else
1392	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1393	fi
1394	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1395	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1396	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1397	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1398	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1399	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1400	if [[ $host != "ibmh" ]]
1401	then
1402	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1403	fi
1404	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1405	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1406
1407	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1408	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1409
1410	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1411	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
1412	if [[ $remote_host = ibmku ]]
1413	then
1414	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1415	fi
1416	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1417
1418	elif [[ $remote_host = nech ]]
1419	then
1420	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1421	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1422	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1423	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1424	echo "#PBS -j o" >> $job_to_send
1425	echo " " >> $job_to_send
1426	echo "set -x" >> $job_to_send
1427	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1428	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1429	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1430	echo "%%END%%" >> $job_to_send
1431
1432	elif [[ $remote_host = necriam ]]
1433	then
1434	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1435	echo "#PBS -q $return_queue" >> $job_to_send
1436	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1437	echo "#PBS -j o" >> $job_to_send
1438	echo " " >> $job_to_send
1439	echo "set -x" >> $job_to_send
1440	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1441	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1442	echo "%%END%%" >> $job_to_send
1443
1444	elif [[ $remote_host = lckyuh ]]
1445	then
1446	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1447	echo "#!/bin/bash" >> $job_to_send
1448	echo "#PJM -L \"node=1\"" >> $job_to_send
1449	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1450	echo "#PJM --no-stging" >> $job_to_send
1451	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1452	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1453	echo "#PJM -j" >> $job_to_send
1454	echo " " >> $job_to_send
1455	echo "export LANG=en_US.UTF-8" >> $job_to_send
1456	echo "set -x" >> $job_to_send
1457	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1458	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1459	echo "%%END%%" >> $job_to_send
1460
1461	elif [[ $remote_host = lckyut ]]
1462	then
1463	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1464	echo "#!/bin/bash" >> $job_to_send
1465	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1466	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1467	echo "#PJM --no-stging" >> $job_to_send
1468	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1469	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1470	echo "#PJM -j" >> $job_to_send
1471	echo " " >> $job_to_send
1472	echo "export LANG=en_US.UTF-8" >> $job_to_send
1473	echo "set -x" >> $job_to_send
1474	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1475	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1476	echo "%%END%%" >> $job_to_send
1477
1478	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1479	then
1480	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1481	echo "#!/bin/bash" >> $job_to_send
1482	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1483	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1484	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1485	echo "#PBS -l feature=data" >> $job_to_send
1486	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1487	echo "#PBS -j oe" >> $job_to_send
1488	echo " " >> $job_to_send
1489	echo ". /usr/share/modules/init/bash" >> $job_to_send
1490	echo "set -x" >> $job_to_send
1491	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1492	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1493	echo "%%END%%" >> $job_to_send
1494
1495	elif [[ $remote_host = lccrayb ]]
1496	then
1497	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1498	echo "#!/bin/bash" >> $job_to_send
1499	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1500	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1501	echo "#PBS -l mppwidth=1" >> $job_to_send
1502	echo "#PBS -l mppnppn=1" >> $job_to_send
1503	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1504	echo "#PBS -j oe" >> $job_to_send
1505	echo " " >> $job_to_send
1506	echo "set -x" >> $job_to_send
1507	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1508	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1509	echo "%%END%%" >> $job_to_send
1510
1511	elif [[ $remote_host = lcflow ]]
1512	then
1513	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1514	echo "#!/bin/bash" >> $job_to_send
1515	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1516	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1517	echo "SGEPREFIX -cwd" >> $job_to_send
1518	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1519	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1520	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1521	echo "SGEPREFIX -j y" >> $job_to_send
1522	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1523	echo " " >> $job_to_send
1524	echo "set -x" >> $job_to_send
1525	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1526	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1527	echo "" >> $job_to_send
1528	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1529	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1530	echo "rm -f scpjob.${kennung}" >> $job_to_send
1531	echo "%%END%%" >> $job_to_send
1532	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1533	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1534
1535	else
1536
1537	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1538	echo "# @\\\$-q $return_queue" >> $job_to_send
1539	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1540	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1541	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1542	then
1543	echo "# @\$-l mpp_p=0" >> $job_to_send
1544	fi
1545	echo '# @\$-lF 10mb' >> $job_to_send
1546	# echo '# @\$-o /dev/null' >> $job_to_send
1547	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1548	echo '# @\\\$-eo' >> $job_to_send
1549	echo " " >> $job_to_send
1550	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1551	then
1552	echo "set +vx" >> $job_to_send
1553	echo ". .profile" >> $job_to_send
1554	fi
1555	echo "set -x" >> $job_to_send
1556	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
1557	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1558	echo "%%END%%" >> $job_to_send
1559	fi
1560
1561	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1562	then
1563	echo "llsubmit scpjob.$kennung" >> $job_to_send
1564	elif [[ $remote_host = lccrayb ]]
1565	then
1566	echo "qsub -q $return_queue scpjob.$kennung" >> $job_to_send
1567	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1568	then
1569	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1570	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1571	echo "msub scpjob.$kennung" >> $job_to_send
1572	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1573	then
1574	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1575	elif [[ $remote_host = t3es ]]
1576	then
1577	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1578	elif [[ $remote_host = lckiaps ]]
1579	then
1580	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1581	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1582	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1583	elif [[ $remote_host = lckyu* ]]
1584	then
1585	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1586	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1587	elif [[ $remote_host = lcflow ]]
1588	then
1589	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1590	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1591	else
1592	echo "$submcom scpjob.$kennung" >> $job_to_send
1593	fi
1594	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1595	then
1596	echo "rm scpjob.$kennung" >> $job_to_send
1597	fi
1598	if [[ $remote_host = nech ]]
1599	then
1600	echo "cd -" >> $job_to_send
1601	fi
1602	else
1603	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1604	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1605	fi
1606	echo "set -x" >> $job_to_send
1607	echo " ' exit" >> $job_to_send
1608	echo "set -x" >> $job_to_send
1609	fi
1610
1611
1612
1613	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1614	cat $file_to_send >> $job_to_send
1615
1616	if [[ $remote_host = ibm ]]
1617	then
1618	echo " " >> $job_to_send
1619	echo "exit" >> $job_to_send
1620	fi
1621
1622	# remove job file
1623	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1624	then
1625	echo " " >> $job_to_send
1626	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1627	fi
1628
1629
1630
1631	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1632	if [[ -z $remote_user ]]
1633	then
1634	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1635	then
1636	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1637	fi
1638	fi
1639
1640
1641
1642	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1643	if [[ $no_submit = false ]]
1644	then
1645	if [[ $remote_host != $local_host ]]
1646	then
1647	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1648	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1649	then
1650	job_catalog_save=$job_catalog
1651	job_catalog=job_queue
1652	elif [[ $remote_host = nech ]]
1653	then
1654	job_catalog_save=$job_catalog
1655	job_catalog=/hpf/b/${remote_user}/job_queue
1656	fi
1657	if [[ $remote_host = nech ]]
1658	then
1659	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1660	# TRANSFERIERT WERDEN
1661	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1662	else
1663	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1664	fi
1665	if [[ $? = 1 ]]
1666	then
1667	locat=scp; exit
1668	fi
1669	if [[ $remote_host = ibms ]]
1670	then
1671	job_catalog=$job_catalog_save
1672	fi
1673	[[ $verify = true ]] && printf "\n >>> finished\n"
1674	else
1675	eval job_catalog=$job_catalog
1676	cp $job_to_send ${job_catalog}/$job_on_remhost
1677	fi
1678
1679
1680
1681	# NQS- BZW. LOADLEVELER-JOB STARTEN
1682	if [[ $remote_host != $local_host ]]
1683	then
1684	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1685
1686	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1687	then
1688	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1689	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1690	elif [[ $remote_host = ibmku ]]
1691	then
1692	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1693	elif [[ $remote_host = lcflow ]]
1694	then
1695	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1696	else
1697	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1698	fi
1699
1700	[[ $verify = true ]] && printf " >>> o.k.\n"
1701	else
1702	cd $job_catalog
1703	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1704	then
1705	eval $submcom $job_on_remhost
1706	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1707	then
1708	chmod u+x $job_on_remhost
1709	eval $submcom $job_on_remhost
1710	elif [[ $local_host = nech ]]
1711	then
1712	if [[ $queue = default ]]
1713	then
1714	eval $submcom $job_on_remhost
1715	else
1716	eval $submcom -q $queue $job_on_remhost
1717	fi
1718	else
1719	qsub $job_on_remhost
1720	fi
1721
1722	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1723	# only at the end of the job.
1724	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1725	then
1726	rm $job_on_remhost
1727	fi
1728	cd - > /dev/null
1729	fi
1730	fi
1731
1732
1733
1734	# ABSCHLUSSARBEITEN
1735
1736	if [[ $no_submit = false ]]
1737	then
1738	rm -f $job_to_send
1739	fi
1740	[[ $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 |