Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1261

Last change on this file since 1261 was 1261, checked in by raasch, 11 years ago
last commit documented
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	#
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1261 2013-12-04 12:48:57Z raasch $
30	#
31	# 1260 2013-12-04 12:48:04Z raasch
32	# jaboticaba admitted
33	#
34	# 1255 2013-11-07 14:43:35Z raasch
35	# further adjustments for lccrayb
36	#
37	# 1224 2013-09-16 07:27:23Z raasch
38	# first adjustments for lccrayb
39	#
40	# 1202 2013-07-10 16:22:07Z witha
41	# adjustments for Forwind cluster (lcflow)
42	#
43	# 1199 2013-07-05 14:52:22Z raasch
44	# adjustments for CSC Helsinki (lccrayf)
45	#
46	# use of cluster/express queue enabled (ibmh)
47	# vinessa added (imuk)
48	#
49	# 1103 2013-02-20 02:15:53Z raasch
50	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
51	# instead typeset -L/R),
52	# further adjustments for lckyuh
53	#
54	# 2013-02-10 01:47:43Z raasch
55	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
56	# and for Forwind cluster (lcflow)
57	#
58	# 1094 2013-02-03 01:52:12Z raasch
59	# new option -P for explicit setting of ssh/scp port,
60	# decalpha parts (yonsei) removed
61	#
62	# 2013-02-02 07:06:13Z raasch
63	# adjustments for Kyushu-University computing center (lckyut - tatara)
64	# old changelog messages removed
65	#
66	# 1046 2012-11-09 14:38:45Z maronga
67	# code put under GPL (PALM 3.9)
68	#
69	# 08/07/94 - Siggi - first version finished
70	# 29/06/94 - Siggi - script development started
71	#--------------------------------------------------------------------------------#
72	# subjob - script for automatic generation and submission of batch-job files
73	# for various batch queuing systems
74	#--------------------------------------------------------------------------------#
75
76
77	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
78	delete_dayfile=false
79	email_notification=none
80	group_number=none
81	locat=normal
82	no_default_queue=none
83	no_submit=false
84	job_catalog="~/job_queue"
85	job_name=none
86	local_user=$LOGNAME
87	node_usage=shared
88	numprocs=0
89	punkte="..........................................................."
90	submcom=qsub
91	queue=default
92	remote_host=none
93	remote_user=""
94	verify=true
95
96	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
97	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
98
99
100
101	# FEHLERBEHANDLUNG
102	# BEI EXIT:
103	trap 'if [[ $locat != normal ]]
104	then
105	case $locat in
106	(option) printf "\n --> available optios can be displayed"
107	printf " by typing:"
108	printf "\n \"subjob ?\" \n";;
109	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
110	(*) printf "\n +++ unknown error"
111	printf "\n please inform S. Raasch!\n"
112	esac
113	[[ -f $job_to_send ]] && rm $job_to_send
114	printf "\n\n+++ SUBJOB killed \n\n"
115	fi' exit
116
117
118	# BEI TERMINAL-BREAK:
119	trap '[[ -f $job_to_send ]] && rm $job_to_send
120	printf "\n\n+++ SUBJOB killed \n\n"
121	exit
122	' 2
123
124
125
126
127	# LOKALEN HOSTNAMEN FESTSTELLEN
128	local_host=$(hostname)
129
130
131
132	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
133	# UEBERHAUPT ZULAESSIG IST
134	# Note: One of the entries for "lck" or "lckordi" always has to be
135	# comment out, because the hostname (node*) is same for both machines
136	case $local_host in
137	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
138	(atmos) local_addres=172.20.25.35; local_host=lcide;;
139	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
140	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
141	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
142	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
143	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
144	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
145	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
146	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
147	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
148	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
149	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
150	(blogin\|bxc) local_addres=130.73.233.1; local_host=lccrayb;;
151	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
152	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
153	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
154	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
155	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
156	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
157	(cs*) local_addres=136.172.44.131; local_host=nech;;
158	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
159	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
160	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
161	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
162	(node*) local_addres=165.132.26.61 local_host=lck;;
163	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
164	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
165	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
166	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
167	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
168	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
169	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
170	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
171	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
172	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
173	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
174	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
175	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
176	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
177	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
178	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
179	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
180	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
181	(jaboticaba) local_addres=150.163.25.181; local_host=lcbr;;
182	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
183	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
184	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
185	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
186	(meller) local_addres=134.106.74.155; local_host=lcfor;;
187	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
188	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
189	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
190	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
191	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
192	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
193	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
194	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
195	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
196	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
197	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
198	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
199	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
200	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
201	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
202	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
203	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
204	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
205	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
206	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
207	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
208	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
209	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
210	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
211	(*) printf "\n +++ \"$local_host\" unknown";
212	printf "\n please contact the PALM group at IMUK";
213	locat=parameter; exit;;
214	esac
215
216
217
218	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
219	remote_host=$local_host
220
221
222
223
224	# PROZEDUROPTIONEN EINLESEN
225	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
226	do
227	case $option in
228	(c) job_catalog=$OPTARG;;
229	(d) delete_dayfile=true;;
230	(D) no_submit=true;;
231	(e) email_notification=$OPTARG;;
232	(g) group_number=$OPTARG;;
233	(h) remote_host=$OPTARG;;
234	(m) memory=$OPTARG;;
235	(n) job_name=$OPTARG;;
236	(N) node_usage=$OPTARG;;
237	(O) threads_per_task=$OPTARG;;
238	(P) scp_port=$OPTARG;;
239	(q) no_default_queue=$OPTARG;;
240	(t) cputime=$OPTARG;;
241	(T) tasks_per_node=$OPTARG;;
242	(u) remote_user=$OPTARG;;
243	(v) verify=false;;
244	(X) numprocs=$OPTARG;;
245	(\?) printf "\n +++ Option $OPTARG unknown \n";
246	locat=option; exit;;
247	esac
248	done
249
250
251	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
252	(( to_shift = $OPTIND - 1 ))
253	shift $to_shift; file_to_send=$1
254
255
256	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
257	if [ "$1" = "?" ]
258	then
259	(printf "\n *** subjob can be called as follows:\n"
260	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
261	printf "\n Description of available options:\n"
262	printf "\n Option Description Default-Value"
263	printf "\n -c job-input- and output-catalog ~/job_queue"
264	printf "\n -d no job-protocol will be created ---"
265	printf "\n -D only the job-file will be created ---"
266	printf "\n -h execution host, available hosts: $remote_host"
267	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
268	printf "\n lckiaps, lctit, nech, necriam, unics"
269	printf "\n -m memory demand per process in MByte ---"
270	printf "\n -n jobname <jobdatei>"
271	printf "\n -O threads per task (for OpenMP usage) 1"
272	printf "\n -P ssh/scp port default port"
273	printf "\n -q job-queue to be used default"
274	printf "\n -t allowed cpu-time in seconds ---"
275	printf "\n -T tasks per node (on parallel hosts) ---"
276	printf "\n -u username on execution host from .netrc"
277	printf "\n -v no prompt for confirmation ---"
278	printf "\n -X # of processors (on parallel hosts) 1"
279	printf "\n "
280	printf "\n The only possible positional parameter is <jobfile>:"
281	printf "\n The complete NQS-job must be provided here."
282	printf "\n <jobfile>=? creates this outline\n\n") \| more
283	exit
284	fi
285
286
287
288	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
289	if [[ "$file_to_send" = "" ]]
290	then
291	printf "\n +++ job-file missing"
292	locat=parameter; exit
293	else
294	if [[ -f $file_to_send ]]
295	then
296	true
297	else
298	printf "\n +++ job-file: "
299	printf "\n $file_to_send"
300	printf "\n does not exist"
301	locat=parameter; exit
302	fi
303	fi
304
305
306
307	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
308	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
309	if [[ $job_name = none ]]
310	then
311	job_name=$file_to_send
312	fi
313	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
314	then
315	printf "\n +++ job-file name: "
316	printf "\n $job_name"
317	printf "\n must not contain \"/\"-characters"
318	locat=parameter; exit
319	fi
320
321
322
323
324	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
325	# ODER WENN HOST NICHT ANGEGEBEN WURDE
326	if [[ $remote_host = none ]]
327	then
328	printf "\n +++ host missing"
329	locat=option; exit
330	else
331	case $remote_host in
332	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
333	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
334	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
335	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
336	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
337	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
338	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/torque/4.2.4.1/bin/qsub;;
339	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
340	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
341	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
342	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
343	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
344	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
345	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
346	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
347	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
348	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
349	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
350	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
351	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
352	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
353	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
354	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
355	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
356	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
357	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
358	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
359	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
360	locat=parameter; exit;;
361	esac
362	fi
363
364
365	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
366	if [[ $no_default_queue != none ]]
367	then
368	error=false
369	ndq=$no_default_queue
370	case $remote_host in
371	(ibm) case $ndq in
372	(p690_express\|p690_standard\|p690_long) error=false;;
373	(*) error=true;;
374	esac;;
375	(ibmh) case $ndq in
376	(cluster\|express) error=false;;
377	(*) error=true;;
378	esac;;
379	(ibmkisti) case $ndq in
380	(class.32plus\|class.1-2\|class.2-32) error=false;;
381	(*) error=true;;
382	esac;;
383	(ibmku) case $ndq in
384	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
385	(*) error=true;;
386	esac;;
387	(ibms) case $ndq in
388	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
389	(*) error=true;;
390	esac;;
391	(ibmy) case $ndq in
392	(parallel) error=false;;
393	(*) error=true;;
394	esac;;
395	(lccrayb) case $ndq in
396	(mpp1q\|mpp1testq\|testq) error=false;;
397	(*) error=true;;
398	esac;;
399	(lccrayf) case $ndq in
400	(usup\|test*\|small\|large) error=false;;
401	(*) error=true;;
402	esac;;
403	(lcflow) case $ndq in
404	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
405	(*) error=true;;
406	esac;;
407	(lckiaps) case $ndq in
408	(express\|normal) error=false;;
409	(*) error=true;;
410	esac;;
411	(lckyoto) case $ndq in
412	(eh\|ph) error=false;;
413	(*) error=true;;
414	esac;;
415	(lckyuh) case $ndq in
416	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
417	(*) error=true;;
418	esac;;
419	(lckyut) case $ndq in
420	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
421	(*) error=true;;
422	esac;;
423	(lcsgib\|lcsgih) case $ndq in
424	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
425	(*) error=true;;
426	esac;;
427	(lctit) case $ndq in
428	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
429	(*) error=true;;
430	esac;;
431	(t3eb) case $ndq in
432	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
433	(*) error=true;;
434	esac;;
435	(necriam) case $ndq in
436	(SP\|SS\|P6) error=false;;
437	(*) error=true;;
438	esac;;
439	(t3eh) case $ndq in
440	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
441	(*) error=true;;
442	esac;;
443	(t3ej2\|t3ej5) case $ndq in
444	(low\|normal\|high) error=false;;
445	(*) error=true;;
446	esac;;
447	(t3es) case $ndq in
448	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
449	(*) error=true;;
450	esac;;
451	(unics) case $ndq in
452	(unics\|ht) error=false;;
453	(*) error=true;;
454	esac;;
455	esac
456	if [[ $error = true ]]
457	then
458	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
459	locat=parameter; exit
460	else
461	queue=$no_default_queue
462	fi
463	fi
464
465
466
467	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
468	# AUFTEILEN
469	done=false
470	while [[ $done = false ]]
471	do
472	if (( $cputime <= 0 ))
473	then
474	printf "\n +++ wrong cpu-time or cpu-time missing"
475	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
476	printf "\n >>> "
477	read cputime 1>/dev/null 2>&1
478	else
479	done=true
480	fi
481	done
482	if [[ $remote_host = nech ]]
483	then
484	if (( tasks_per_node != 0 ))
485	then
486	(( cputime = cputime * tasks_per_node ))
487	elif [[ $numprocs != 0 ]]
488	then
489	(( cputime = cputime * numprocs ))
490	fi
491	fi
492	(( stunden = cputime / 3600 ))
493	(( resttime = cputime - stunden * 3600 ))
494	(( minuten = resttime / 60 ))
495	(( sekunden = resttime - minuten * 60 ))
496	timestring=${stunden}:${minuten}:${sekunden}
497
498
499
500	# PRUEFEN DER KERNSPEICHERANFORDERUNG
501	done=false
502	while [[ $done = false ]]
503	do
504	if (( memory <= 0 ))
505	then
506	printf "\n +++ wrong memory demand or memory demand missing"
507	printf "\n >>> Please type memory in MByte per process as INTEGER:"
508	printf "\n >>> "
509	read memory 1>/dev/null 2>&1
510	else
511	done=true
512	fi
513	done
514
515	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
516	then
517	if (( tasks_per_node != 0 ))
518	then
519	(( Memory = memory * tasks_per_node / 1000 ))
520	elif [[ $numprocs != 0 ]]
521	then
522	(( Memory = memory * numprocs / 1000 ))
523	else
524	(( Memory = memory / 1000 ))
525	fi
526	elif [[ $remote_host = lctit ]]
527	then
528	(( Memory = memory * tasks_per_node / 1000 ))
529	fi
530
531
532	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
533	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
534	then
535	(( memory = memory * threads_per_task ))
536	fi
537
538
539	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
540	if (( tasks_per_node != 0 ))
541	then
542	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
543	fi
544
545
546	# Calculate number of processes per node
547	(( processes_per_node = tasks_per_node * threads_per_task ))
548
549
550	# Calculate number of MPI tasks
551	(( mpi_tasks = numprocs / threads_per_task ))
552
553
554	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
555	if [[ "$scp_port" != "" ]]
556	then
557	PORTOPT="-P $scp_port"
558	SSH_PORTOPT="-p $scp_port"
559	fi
560
561
562	# HEADER-AUSGABE
563	if [[ $verify = true ]]
564	then
565	printf "\n\n"
566	printf "#--------------------------------------------------------------# \n"
567	spalte1=SUBJOB;spalte2=$(date)
568	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
569	printf "\| \| \n"
570	printf "\| values of parameters/options: \| \n"
571	spalte1=$(echo local_host$punkte \| cut -c-20)
572	spalte2=$punkte$local_host
573	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
574	spalte1=$(echo remote_host$punkte \| cut -c-20)
575	spalte2=$punkte$remote_host
576	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
577	spalte1=$(echo queue$punkte \| cut -c-20)
578	spalte2=$punkte$queue
579	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
580	spalte1=$(echo memory$punkte \| cut -c-20)
581	spalte2="$punkte$memory mb"
582	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
583	spalte1=$(echo cputime$punkte \| cut -c-20)
584	spalte2="$punkte$cputime sec"
585	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
586	spalte1=$(echo job_name$punkte \| cut -c-20)
587	spalte2="$punkte$job_name"
588	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
589	printf "#--------------------------------------------------------------# \n\n"
590
591
592	# KONTROLLABFRAGE, OB ALLES O.K.
593	antwort="dummy"
594	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
595	do
596	read antwort?" >>> continue (y/n) ? "
597	done
598	if [[ $antwort = n \|\| $antwort = N ]]
599	then
600	locat=verify; exit
601	fi
602	printf "\n"
603	fi
604
605
606
607	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
608	kennung=$RANDOM
609	job_on_remhost=${job_name}_${kennung}_$local_host
610	job_to_send=job_to_send_$kennung
611	if [[ $delete_dayfile = false ]]
612	then
613	remote_dayfile=${local_host}_${job_name}_result_$kennung
614	local_dayfile=${remote_host}_${job_name}
615	else
616	remote_dayfile=/dev/null
617	fi
618
619
620
621	# Generate the batch job scripts (qsub/msub/LoadLeveler)
622	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
623	then
624
625	# General LoadLeveler settings
626	execute_in_shell="#!/bin/ksh"
627	use_shell="# @ shell = /bin/ksh"
628	consumable_memory="ConsumableMemory($memory mb)"
629	class="# @ class = $queue"
630	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
631	network_to_use="# @ network.mpi = sn_all,shared,us"
632	data_limit="# @ data_limit = 1.76gb"
633	image_size="# @ image_size = 50"
634	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
635
636	if [[ $email_notification = none ]]
637	then
638	notify_user=""
639	else
640	notify_user="# @ notify_user = $email_notification"
641	if [[ $delete_dayfile = true ]]
642	then
643	notification='# @ notification = never'
644	fi
645	fi
646
647	if [[ $remote_host = ibmh ]]
648	then
649	data_limit=""
650	network_to_use=""
651	class="# @ class = $queue"
652	environment=""
653	rset="# @ rset = RSET_MCM_AFFINITY"
654	task_affinity="# @ task_affinity = core(1)"
655	elif [[ $remote_host = ibmkisti ]]
656	then
657	network_to_use="# @ network.MPI = sn_all,shared,US"
658	wall_clock_limit="# @ wall_clock_limit = $timestring"
659	if [[ $threads_per_task = 1 ]]
660	then
661	rset="# @ rset = RSET_MCM_AFFINITY"
662	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
663	fi
664	environment=""
665	use_shell=""
666	data_limit=""
667	image_size=""
668	elif [[ $remote_host = ibmku ]]
669	then
670	execute_in_shell="#!/usr/bin/ksh"
671	use_shell="# @ shell = /usr/bin/ksh"
672	consumable_memory=""
673	environment=""
674	network_to_use="# @ network.mpi = sn_all,shared,us"
675	data_limit=""
676	image_size=""
677	elif [[ $remote_host = ibms ]]
678	then
679	network_to_use="# @ network.mpi = csss,shared,us"
680	elif [[ $remote_host = ibmy ]]
681	then
682	consumable_memory=""
683	network_to_use=""
684	fi
685
686	cat > $job_to_send << %%END%%
687	$execute_in_shell
688	$use_shell
689
690	# @ job_type = parallel
691	# @ job_name = $job_name
692	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
693	# @ output = $remote_dayfile
694	# @ error = $remote_dayfile
695	$wall_clock_limit
696	$image_size
697	$class
698	$environment
699	$network_to_use
700	$data_limit
701	$rset
702	$mcm_affinity_options
703	$task_affinity
704	$notification
705	$notify_user
706
707	%%END%%
708
709	if (( nodes > 0 ))
710	then
711
712	if [[ $remote_host != ibmkisti ]]
713	then
714
715	cat >> $job_to_send << %%END%%
716	# @ node = $nodes
717	# @ tasks_per_node = $processes_per_node
718	# @ node_usage = $node_usage
719	# @ queue
720
721	%%END%%
722
723	else
724
725	cat >> $job_to_send << %%END%%
726	# @ total_tasks = $mpi_tasks
727	# @ blocking = unlimited
728	# @ queue
729
730	%%END%%
731
732	fi
733
734	else
735
736	if [[ $remote_host != ibmy ]]
737	then
738
739	cat >> $job_to_send << %%END%%
740	# @ blocking = unlimited
741	# @ total_tasks = $numprocs
742	# @ node_usage = $node_usage
743	# @ queue
744
745	%%END%%
746
747	else
748
749	cat >> $job_to_send << %%END%%
750	# @ node = 1
751	# @ total_tasks = $numprocs
752	# @ queue
753
754	%%END%%
755
756	fi
757
758	fi
759
760	# workaround because of silly job filter on ibmkisti
761	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
762	then
763	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
764	fi
765
766	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
767	then
768
769	cat > $job_to_send << %%END%%
770	#!/bin/ksh
771
772	# @ job_type = serial
773	# @ node_usage = $node_usage
774	# @ job_name = palm
775	# @ wall_clock_limit = ${timestring},$timestring
776	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
777	# @ output = $remote_dayfile
778	# @ error = $remote_dayfile
779	$class
780	$notification
781
782	# @ queue
783
784	%%END%%
785
786	elif [[ $remote_host = lccrayb ]]
787	then
788
789	if [[ $numprocs != 0 ]]
790	then
791	cat > $job_to_send << %%END%%
792	#!/bin/bash
793	#PBS -N $job_name
794	#PBS -l walltime=$timestring
795	#PBS -l mppwidth=$numprocs
796	#PBS -l mppnppn=$processes_per_node
797	#PBS -o $remote_dayfile
798	#PBS -j oe
799	#PBS -q $queue
800
801	$module_calls
802
803	%%END%%
804
805	else
806
807	continue
808
809	fi
810
811	elif [[ $remote_host = lccrayf ]]
812	then
813
814	if [[ $numprocs != 0 ]]
815	then
816	cat > $job_to_send << %%END%%
817	#!/bin/bash -l
818	#SBATCH -J $job_name
819	#SBATCH -t $timestring
820	#SBATCH -N $nodes
821	#SBATCH --ntasks-per-node=$processes_per_node
822	#SBATCH -p $queue
823	#SBATCH -o $remote_dayfile
824	#SBATCH -e $remote_dayfile
825
826	$init_cmds
827	$module_calls
828
829	%%END%%
830
831	else
832	cat > $job_to_send << %%END%%
833	#!/bin/bash -l
834	#SBATCH -J $job_name
835	#SBATCH -t $timestring
836	#SBATCH -l ncpus=1
837	#SBATCH -l pmem=${memory}mb
838	#SBATCH -m abe
839	#SBATCH -o $remote_dayfile
840	#SBATCH -e $remote_dayfile
841
842	$init_cmds
843	$module_calls
844
845	%%END%%
846
847	fi
848
849	elif [[ $remote_host = lcfimm ]]
850	then
851
852	if [[ $numprocs != 0 ]]
853	then
854	cat > $job_to_send << %%END%%
855	#!/bin/ksh
856	#PBS -N $job_name
857	#PBS -A $project_account
858	#PBS -l walltime=$timestring
859	#PBS -l nodes=${nodes}:ppn=$processes_per_node
860	#PBS -l pmem=${memory}mb
861	#PBS -m abe
862	#PBS -o $remote_dayfile
863	#PBS -j oe
864	mpd &
865
866	%%END%%
867
868	else
869	cat > $job_to_send << %%END%%
870	#!/bin/ksh
871	#PBS -N $job_name
872	#PBS -A $project_account
873	#PBS -l walltime=$timestring
874	#PBS -l ncpus=1
875	#PBS -l pmem=${memory}mb
876	#PBS -m abe
877	#PBS -o $remote_dayfile
878	#PBS -j oe
879
880	%%END%%
881
882	fi
883
884	elif [[ $remote_host = lcflow ]]
885	then
886	if [ $memory -gt 1800 ]; then
887	use_himem=""
888	else
889	use_himem="#"
890	fi
891	if [[ $numprocs != 0 ]]
892	then
893	pe_set="#$ -pe impi41 $numprocs"
894	else
895	pe_set="#$ -pe impi41 1"
896	fi
897	if [[ $queue = default ]]
898	then
899	queue_set=""
900	else
901	queue_set="#$ -q $queue"
902	fi
903	[[ "$disc_space" = "" ]] && disc_space=50
904
905	cat > $job_to_send << %%END%%
906	#!/bin/bash
907	#$ -S /bin/bash
908	#$ -N $job_name
909	#$ -cwd
910	#$ -l h_rt=$timestring
911	#$ -l h_vmem=${memory}M
912	#$ -o $remote_dayfile
913	#$ -j y
914	$pe_set
915	#$ -R y
916	${use_himem}#$ -l highmem=true
917	#$ -l h_fsize=${disc_space}G
918	$queue_set
919
920	%%END%%
921
922	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
923	then
924
925	if [[ $numprocs != 0 ]]
926	then
927	cat > $job_to_send << %%END%%
928	#!/bin/ksh
929	#PBS -N $job_name
930	#PBS -l walltime=$timestring
931	#PBS -l ncpus=$numprocs
932	#PBS -l pmem=${memory}mb
933	#PBS -o $remote_dayfile
934	#PBS -l nodes=$nodes:ppn=${processes_per_node}
935	#PBS -j oe
936
937	mpd &
938
939	%%END%%
940
941	else
942	cat > $job_to_send << %%END%%
943	#!/bin/ksh
944	#PBS -N $job_name
945	#PBS -l walltime=$timestring
946	#PBS -l ncpus=1
947	#PBS -l pmem=${memory}mb
948	#PBS -o $remote_dayfile
949	#PBS -j oe
950
951	%%END%%
952
953	fi
954
955	elif [[ $remote_host = lckiaps ]]
956	then
957
958	if [[ $numprocs != 0 ]]
959	then
960	cat > $job_to_send << %%END%%
961	#!/bin/ksh
962	#PBS -N $job_name
963	#PBS -l walltime=$timestring
964	#PBS -l select=1:ncpus=$numprocs
965	#PBS -l pmem=${memory}mb
966	#PBS -q $queue
967	#PBS -o $remote_dayfile
968	#PBS -j oe
969	#PBS -V
970
971	%%END%%
972
973	else
974	cat > $job_to_send << %%END%%
975	#!/bin/ksh
976	#PBS -N $job_name
977	#PBS -l walltime=$timestring
978	#PBS -l ncpus=1
979	#PBS -l pmem=${memory}mb
980	#PBS -o $remote_dayfile
981	#PBS -j oe
982
983	%%END%%
984
985	fi
986
987	elif [[ $remote_host = lcyon ]]
988	then
989
990	if [[ $numprocs != 0 ]]
991	then
992	cat > $job_to_send << %%END%%
993	#!/bin/ksh
994	#PBS -N $job_name
995	#PBS -l walltime=$timestring
996	#PBS -l ncpus=$numprocs
997	#PBS -l pmem=${memory}mb
998	#PBS -o $remote_dayfile
999	#PBS -j oe
1000
1001	%%END%%
1002
1003	else
1004	cat > $job_to_send << %%END%%
1005	#!/bin/ksh
1006	#PBS -N $job_name
1007	#PBS -l walltime=$timestring
1008	#PBS -l ncpus=1
1009	#PBS -l pmem=${memory}mb
1010	#PBS -o $remote_dayfile
1011	#PBS -j oe
1012
1013	%%END%%
1014
1015	fi
1016
1017	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1018	then
1019
1020	if [[ "$sgi_feature" != "" ]]
1021	then
1022	feature_directive="#PBS -l feature=$sgi_feature"
1023	else
1024	feature_directive=""
1025	fi
1026
1027	if [[ $queue = dataq \|\| $queue = permq ]]
1028	then
1029	feature_directive="#PBS -l feature=data"
1030	fi
1031
1032	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
1033	then
1034	queue_directive="#PBS -q $queue"
1035	else
1036	queue_directive=""
1037	fi
1038
1039	if [[ $email_notification = none ]]
1040	then
1041	email_directive=""
1042	else
1043	email_directive="#PBS -M $email_notification"
1044	fi
1045
1046	if [[ $numprocs != 0 ]]
1047	then
1048	cat > $job_to_send << %%END%%
1049	#!/bin/bash
1050	#PBS -N $job_name
1051	#PBS -l walltime=$timestring
1052	#PBS -l nodes=$nodes:ppn=${processes_per_node}
1053	#PBS -l naccesspolicy=$node_usage
1054	#PBS -o $remote_dayfile
1055	#PBS -j oe
1056	$feature_directive
1057	$queue_directive
1058	$email_directive
1059
1060	eval \`/sw/swdist/bin/modulesinit\`
1061	#. /usr/share/modules/init/bash
1062	$init_cmds
1063	$module_calls
1064
1065	echo ld_library_path=\$LD_LIBRARY_PATH
1066
1067	%%END%%
1068
1069	else
1070	cat > $job_to_send << %%END%%
1071	#PBS -S /bin/bash
1072	#PBS -N $job_name
1073	#PBS -l walltime=$timestring
1074	#PBS -l ncpus=1
1075	#PBS -o $remote_dayfile
1076	#PBS -j oe
1077	$queue_directive
1078	$email_directive
1079
1080	eval \`/sw/swdist/bin/modulesinit\`
1081	#. /usr/share/modules/init/bash
1082	$init_cmds
1083	$module_calls
1084
1085	%%END%%
1086
1087	fi
1088
1089	elif [[ $remote_host = lcxe6 ]]
1090	then
1091
1092	if [[ $numprocs != 0 ]]
1093	then
1094	cat > $job_to_send << %%END%%
1095	#!/bin/ksh
1096	#PBS -S /bin/ksh
1097	#PBS -N $job_name
1098	#PBS -A $project_account
1099	#PBS -j oe
1100	#PBS -l walltime=$timestring
1101	#PBS -l mppwidth=${numprocs}
1102	#PBS -l mppnppn=${processes_per_node}
1103	#PBS -m abe
1104	#PBS -o $remote_dayfile
1105	$email_directive
1106
1107	$init_cmds
1108	$module_calls
1109
1110	%%END%%
1111
1112	else
1113	cat > $job_to_send << %%END%%
1114	#!/bin/ksh
1115	#PBS -S /bin/ksh
1116	#PBS -N $job_name
1117	#PBS -A $project_account
1118	#PBS -j oe
1119	#PBS -l walltime=$timestring
1120	#PBS -l ncpus=1
1121	#PBS -l pmem=${memory}mb
1122	#PBS -m abe
1123	$email_directive
1124	#PBS -o $remote_dayfile
1125
1126	$init_cmds
1127	$module_calls
1128
1129	%%END%%
1130
1131	fi
1132
1133	elif [[ $remote_host = lckyoto ]]
1134	then
1135
1136	cat > $job_to_send << %%END%%
1137	#!/bin/ksh
1138	# @\$-o $remote_dayfile
1139	# @\$-eo -oi
1140	# @\$-lP 16
1141	# @\$-lp 1
1142	# @\$-lm 28gb -llm unlimited -ls unlimited
1143	# @\$-q $queue
1144	# @\$-Pvn abs_pack
1145	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1146	#. /thin/local/etc/setprofile/intel-11.0.sh
1147	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1148	. ~/.myprofile
1149	#. /home2/t/t51254/palm/current_version/myprofile
1150	#. /thin/apps/pgi/mpi.sh
1151	#
1152	env
1153	#
1154	set -x
1155
1156	%%END%%
1157
1158	elif [[ $remote_host = lcxt5m ]]
1159	then
1160
1161	if [[ $numprocs != 0 ]]
1162	then
1163	cat > $job_to_send << %%END%%
1164	#!/bin/ksh
1165	#PBS -S /bin/ksh
1166	#PBS -N $job_name
1167	#PBS -j oe
1168	#PBS -l walltime=$timestring
1169	#PBS -l mppwidth=${numprocs}
1170	#PBS -l mppnppn=${processes_per_node}
1171	#PBS -m abe
1172	#PBS -o $remote_dayfile
1173
1174	$init_cmds
1175	$module_calls
1176
1177	%%END%%
1178
1179	else
1180	cat > $job_to_send << %%END%%
1181	#!/bin/ksh
1182	#PBS -S /bin/ksh
1183	#PBS -N $job_name
1184	#PBS -j oe
1185	#PBS -l walltime=$timestring
1186	#PBS -l ncpus=1
1187	#PBS -l pmem=${memory}mb
1188	#PBS -m abe
1189	#PBS -o $remote_dayfile
1190
1191	$init_cmds
1192	$module_calls
1193
1194	%%END%%
1195
1196	fi
1197
1198	elif [[ $remote_host = lckyuh ]]
1199	then
1200	cat > $job_to_send << %%END%%
1201	#!/bin/bash
1202	#PJM -L "rscgrp=$queue"
1203	#PJM -L "node=$nodes"
1204	#PJM --mpi "proc=$numprocs"
1205	#PJM -L "elapse=$timestring"
1206	#PJM -o $remote_dayfile
1207	#PJM -j
1208	#PJM -X
1209	#PJM --no-stging
1210
1211	export LANG=en_US.UTF-8
1212	%%END%%
1213
1214	elif [[ $remote_host = lckyut ]]
1215	then
1216	cat > $job_to_send << %%END%%
1217	#!/bin/bash
1218	#PJM -L "rscgrp=$queue"
1219	#PJM -L "vnode=$numprocs"
1220	#PJM -L "vnode-core=1"
1221	#PJM -L "elapse=$timestring"
1222	#PJM --mpi proc=$numprocs
1223	#PJM -o $remote_dayfile
1224	#PJM -j
1225	#PJM -X
1226	#PJM --no-stging
1227
1228	export LANG=en_US.UTF-8
1229	%%END%%
1230
1231	elif [[ $remote_host = nech ]]
1232	then
1233
1234	if (( nodes > 1 ))
1235	then
1236	cat > $job_to_send << %%END%%
1237	#!/bin/ksh
1238	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1239	#PBS -l ${qsubmem}=${Memory}gb
1240	#PBS -b $nodes
1241	#PBS -o $remote_dayfile
1242	#PBS -N palm
1243	#PBS -j o
1244	#PBS -T mpisx
1245
1246	%%END%%
1247
1248	elif [[ $numprocs != 0 ]]
1249	then
1250	cat > $job_to_send << %%END%%
1251	#!/bin/ksh
1252	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1253	#PBS -l ${qsubmem}=${Memory}gb
1254	#PBS -o $remote_dayfile
1255	#PBS -N palm
1256	#PBS -j o
1257
1258	%%END%%
1259
1260	else
1261	cat > $job_to_send << %%END%%
1262	#!/bin/ksh
1263	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1264	#PBS -o $remote_dayfile
1265	#PBS -j o
1266
1267	%%END%%
1268
1269	fi
1270
1271	elif [[ $remote_host = necriam ]]
1272	then
1273
1274	if (( nodes > 1 ))
1275	then
1276	cat > $job_to_send << %%END%%
1277	#!/bin/ksh
1278	#PBS -b $nodes
1279	#PBS -o $remote_dayfile
1280	#PBS -N $job_name
1281	#PBS -j o
1282	#PBS -v MPIPROGINV=YES
1283
1284	%%END%%
1285
1286	elif [[ $numprocs != 0 ]]
1287	then
1288	cat > $job_to_send << %%END%%
1289	#!/bin/ksh
1290	#PBS -q ${queue}
1291	#PBS -o $remote_dayfile
1292	#PBS -N $job_name
1293	#PBS -j o
1294	#PBS -v MPIPROGINV=YES
1295
1296	%%END%%
1297
1298	else
1299	cat > $job_to_send << %%END%%
1300	#!/bin/ksh
1301	#PBS -o $remote_dayfile
1302	#PBS -j o
1303
1304	%%END%%
1305
1306	fi
1307
1308	elif [[ $remote_host = lctit ]]
1309	then
1310	cat > $job_to_send << %%END%%
1311	#!/bin/ksh
1312	$init_cmds
1313	$module_calls
1314
1315	%%END%%
1316
1317	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1318	if [[ $tasks_per_node != $processes_per_node ]]
1319	then
1320	submcom="$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 $queue "
1321	else
1322	submcom="$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 "
1323	fi
1324
1325	else
1326
1327	cat > $job_to_send << %%END%%
1328	# @\$-q ${queue}
1329	# @\$-l${qsubtime} $timestring
1330	# @\$-l${qsubmem} ${memory}mb
1331	# @\$-o $remote_dayfile
1332	# @\$-eo
1333
1334	%%END%%
1335
1336	fi
1337
1338
1339	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1340	# BEI EXIT VERANLASSEN
1341	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1342	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1343	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1344	then
1345	echo "set +vx" >> $job_to_send
1346	echo "trap '" >> $job_to_send
1347	echo "set +vx" >> $job_to_send
1348	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* ]]
1349	then
1350	if [[ $remote_host = ibmh ]]
1351	then
1352	return_queue=c1
1353	elif [[ $remote_host = ibmkisti ]]
1354	then
1355	return_queue=class.1-2
1356	elif [[ $remote_host = ibmku ]]
1357	then
1358	return_queue=sdbg2
1359	elif [[ $remote_host = ibms ]]
1360	then
1361	return_queue=p_normal
1362	elif [[ $remote_host = ibmy ]]
1363	then
1364	return_queue=serial
1365	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1366	then
1367	return_queue=serialq
1368	elif [[ $remote_host = necriam ]]
1369	then
1370	return_queue=SP
1371	elif [[ $remote_host = lccrayb ]]
1372	then
1373	return_queue=dataq
1374	elif [[ $remote_host = lckiaps ]]
1375	then
1376	return_queue=express
1377	elif [[ $remote_host = lckyuh ]]
1378	then
1379	return_queue=cx-single
1380	elif [[ $remote_host = lckyut ]]
1381	then
1382	return_queue=cx-single
1383	else
1384	return_queue=unknown
1385	fi
1386
1387	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1388	then
1389
1390	if [[ $remote_host = ibmku ]]
1391	then
1392	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1393	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1394	else
1395	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1396	fi
1397	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1398	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1399	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1400	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1401	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1402	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1403	if [[ $host != "ibmh" ]]
1404	then
1405	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1406	fi
1407	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1408	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1409
1410	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1411	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1412
1413	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1414	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
1415	if [[ $remote_host = ibmku ]]
1416	then
1417	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1418	fi
1419	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1420
1421	elif [[ $remote_host = nech ]]
1422	then
1423	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1424	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1425	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1426	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1427	echo "#PBS -j o" >> $job_to_send
1428	echo " " >> $job_to_send
1429	echo "set -x" >> $job_to_send
1430	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1431	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1432	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1433	echo "%%END%%" >> $job_to_send
1434
1435	elif [[ $remote_host = necriam ]]
1436	then
1437	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1438	echo "#PBS -q $return_queue" >> $job_to_send
1439	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1440	echo "#PBS -j o" >> $job_to_send
1441	echo " " >> $job_to_send
1442	echo "set -x" >> $job_to_send
1443	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1444	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1445	echo "%%END%%" >> $job_to_send
1446
1447	elif [[ $remote_host = lckyuh ]]
1448	then
1449	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1450	echo "#!/bin/bash" >> $job_to_send
1451	echo "#PJM -L \"node=1\"" >> $job_to_send
1452	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1453	echo "#PJM --no-stging" >> $job_to_send
1454	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1455	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1456	echo "#PJM -j" >> $job_to_send
1457	echo " " >> $job_to_send
1458	echo "export LANG=en_US.UTF-8" >> $job_to_send
1459	echo "set -x" >> $job_to_send
1460	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1461	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1462	echo "%%END%%" >> $job_to_send
1463
1464	elif [[ $remote_host = lckyut ]]
1465	then
1466	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1467	echo "#!/bin/bash" >> $job_to_send
1468	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1469	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1470	echo "#PJM --no-stging" >> $job_to_send
1471	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1472	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1473	echo "#PJM -j" >> $job_to_send
1474	echo " " >> $job_to_send
1475	echo "export LANG=en_US.UTF-8" >> $job_to_send
1476	echo "set -x" >> $job_to_send
1477	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1478	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1479	echo "%%END%%" >> $job_to_send
1480
1481	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1482	then
1483	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1484	echo "#!/bin/bash" >> $job_to_send
1485	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1486	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1487	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1488	echo "#PBS -l feature=data" >> $job_to_send
1489	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1490	echo "#PBS -j oe" >> $job_to_send
1491	echo " " >> $job_to_send
1492	echo ". /usr/share/modules/init/bash" >> $job_to_send
1493	echo "set -x" >> $job_to_send
1494	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1495	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1496	echo "%%END%%" >> $job_to_send
1497
1498	elif [[ $remote_host = lccrayb ]]
1499	then
1500	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1501	echo "#!/bin/bash" >> $job_to_send
1502	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1503	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1504	echo "#PBS -l mppwidth=1" >> $job_to_send
1505	echo "#PBS -l mppnppn=1" >> $job_to_send
1506	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1507	echo "#PBS -j oe" >> $job_to_send
1508	echo " " >> $job_to_send
1509	echo "set -x" >> $job_to_send
1510	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1511	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1512	echo "%%END%%" >> $job_to_send
1513
1514	elif [[ $remote_host = lcflow ]]
1515	then
1516	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1517	echo "#!/bin/bash" >> $job_to_send
1518	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1519	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1520	echo "SGEPREFIX -cwd" >> $job_to_send
1521	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1522	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1523	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1524	echo "SGEPREFIX -j y" >> $job_to_send
1525	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1526	echo " " >> $job_to_send
1527	echo "set -x" >> $job_to_send
1528	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1529	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1530	echo "" >> $job_to_send
1531	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1532	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1533	echo "rm -f scpjob.${kennung}" >> $job_to_send
1534	echo "%%END%%" >> $job_to_send
1535	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1536	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1537
1538	else
1539
1540	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1541	echo "# @\\\$-q $return_queue" >> $job_to_send
1542	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1543	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1544	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1545	then
1546	echo "# @\$-l mpp_p=0" >> $job_to_send
1547	fi
1548	echo '# @\$-lF 10mb' >> $job_to_send
1549	# echo '# @\$-o /dev/null' >> $job_to_send
1550	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1551	echo '# @\\\$-eo' >> $job_to_send
1552	echo " " >> $job_to_send
1553	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1554	then
1555	echo "set +vx" >> $job_to_send
1556	echo ". .profile" >> $job_to_send
1557	fi
1558	echo "set -x" >> $job_to_send
1559	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
1560	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1561	echo "%%END%%" >> $job_to_send
1562	fi
1563
1564	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1565	then
1566	echo "llsubmit scpjob.$kennung" >> $job_to_send
1567	elif [[ $remote_host = lccrayb ]]
1568	then
1569	echo "qsub -q $return_queue scpjob.$kennung" >> $job_to_send
1570	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1571	then
1572	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1573	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1574	echo "msub scpjob.$kennung" >> $job_to_send
1575	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1576	then
1577	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1578	elif [[ $remote_host = t3es ]]
1579	then
1580	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1581	elif [[ $remote_host = lckiaps ]]
1582	then
1583	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1584	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1585	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1586	elif [[ $remote_host = lckyu* ]]
1587	then
1588	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1589	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1590	elif [[ $remote_host = lcflow ]]
1591	then
1592	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1593	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1594	else
1595	echo "$submcom scpjob.$kennung" >> $job_to_send
1596	fi
1597	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1598	then
1599	echo "rm scpjob.$kennung" >> $job_to_send
1600	fi
1601	if [[ $remote_host = nech ]]
1602	then
1603	echo "cd -" >> $job_to_send
1604	fi
1605	else
1606	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1607	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1608	fi
1609	echo "set -x" >> $job_to_send
1610	echo " ' exit" >> $job_to_send
1611	echo "set -x" >> $job_to_send
1612	fi
1613
1614
1615
1616	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1617	cat $file_to_send >> $job_to_send
1618
1619	if [[ $remote_host = ibm ]]
1620	then
1621	echo " " >> $job_to_send
1622	echo "exit" >> $job_to_send
1623	fi
1624
1625	# remove job file
1626	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1627	then
1628	echo " " >> $job_to_send
1629	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1630	fi
1631
1632
1633
1634	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1635	if [[ -z $remote_user ]]
1636	then
1637	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1638	then
1639	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1640	fi
1641	fi
1642
1643
1644
1645	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1646	if [[ $no_submit = false ]]
1647	then
1648	if [[ $remote_host != $local_host ]]
1649	then
1650	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1651	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1652	then
1653	job_catalog_save=$job_catalog
1654	job_catalog=job_queue
1655	elif [[ $remote_host = nech ]]
1656	then
1657	job_catalog_save=$job_catalog
1658	job_catalog=/hpf/b/${remote_user}/job_queue
1659	fi
1660	if [[ $remote_host = nech ]]
1661	then
1662	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1663	# TRANSFERIERT WERDEN
1664	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1665	else
1666	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1667	fi
1668	if [[ $? = 1 ]]
1669	then
1670	locat=scp; exit
1671	fi
1672	if [[ $remote_host = ibms ]]
1673	then
1674	job_catalog=$job_catalog_save
1675	fi
1676	[[ $verify = true ]] && printf "\n >>> finished\n"
1677	else
1678	eval job_catalog=$job_catalog
1679	cp $job_to_send ${job_catalog}/$job_on_remhost
1680	fi
1681
1682
1683
1684	# NQS- BZW. LOADLEVELER-JOB STARTEN
1685	if [[ $remote_host != $local_host ]]
1686	then
1687	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1688
1689	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1690	then
1691	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1692	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1693	elif [[ $remote_host = ibmku ]]
1694	then
1695	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1696	elif [[ $remote_host = lcflow ]]
1697	then
1698	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1699	else
1700	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1701	fi
1702
1703	[[ $verify = true ]] && printf " >>> o.k.\n"
1704	else
1705	cd $job_catalog
1706	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1707	then
1708	eval $submcom $job_on_remhost
1709	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1710	then
1711	chmod u+x $job_on_remhost
1712	eval $submcom $job_on_remhost
1713	elif [[ $local_host = nech ]]
1714	then
1715	if [[ $queue = default ]]
1716	then
1717	eval $submcom $job_on_remhost
1718	else
1719	eval $submcom -q $queue $job_on_remhost
1720	fi
1721	else
1722	qsub $job_on_remhost
1723	fi
1724
1725	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1726	# only at the end of the job.
1727	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1728	then
1729	rm $job_on_remhost
1730	fi
1731	cd - > /dev/null
1732	fi
1733	fi
1734
1735
1736
1737	# ABSCHLUSSARBEITEN
1738
1739	if [[ $no_submit = false ]]
1740	then
1741	rm -f $job_to_send
1742	fi
1743	[[ $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 |