Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1266

Last change on this file since 1266 was 1266, checked in by heinze, 10 years ago
adjustment for lccrayb (use msub instead of qsub)
Property svn:keywords set to `Id Rev`
File size: 58.4 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	# further adjustments for lccrayb (use msub instead of qsub)
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1266 2013-12-11 12:07:34Z heinze $
30	#
31	# 1264 2013-12-09 12:46:09Z fricke
32	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
33	#
34	# 1262 2013-12-09 10:57:20Z fricke
35	# further adjustments for lccrayb
36	#
37	# 1260 2013-12-04 12:48:04Z raasch
38	# jaboticaba admitted
39	#
40	# 1255 2013-11-07 14:43:35Z raasch
41	# further adjustments for lccrayb
42	#
43	# 1224 2013-09-16 07:27:23Z raasch
44	# first adjustments for lccrayb
45	#
46	# 1202 2013-07-10 16:22:07Z witha
47	# adjustments for Forwind cluster (lcflow)
48	#
49	# 1199 2013-07-05 14:52:22Z raasch
50	# adjustments for CSC Helsinki (lccrayf)
51	#
52	# use of cluster/express queue enabled (ibmh)
53	# vinessa added (imuk)
54	#
55	# 1103 2013-02-20 02:15:53Z raasch
56	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
57	# instead typeset -L/R),
58	# further adjustments for lckyuh
59	#
60	# 2013-02-10 01:47:43Z raasch
61	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
62	# and for Forwind cluster (lcflow)
63	#
64	# 1094 2013-02-03 01:52:12Z raasch
65	# new option -P for explicit setting of ssh/scp port,
66	# decalpha parts (yonsei) removed
67	#
68	# 2013-02-02 07:06:13Z raasch
69	# adjustments for Kyushu-University computing center (lckyut - tatara)
70	# old changelog messages removed
71	#
72	# 1046 2012-11-09 14:38:45Z maronga
73	# code put under GPL (PALM 3.9)
74	#
75	# 08/07/94 - Siggi - first version finished
76	# 29/06/94 - Siggi - script development started
77	#--------------------------------------------------------------------------------#
78	# subjob - script for automatic generation and submission of batch-job files
79	# for various batch queuing systems
80	#--------------------------------------------------------------------------------#
81
82
83	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
84	delete_dayfile=false
85	email_notification=none
86	group_number=none
87	locat=normal
88	no_default_queue=none
89	no_submit=false
90	job_catalog="~/job_queue"
91	job_name=none
92	local_user=$LOGNAME
93	node_usage=shared
94	numprocs=0
95	punkte="..........................................................."
96	submcom=qsub
97	queue=default
98	remote_host=none
99	remote_user=""
100	verify=true
101
102	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
103	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
104
105
106
107	# FEHLERBEHANDLUNG
108	# BEI EXIT:
109	trap 'if [[ $locat != normal ]]
110	then
111	case $locat in
112	(option) printf "\n --> available optios can be displayed"
113	printf " by typing:"
114	printf "\n \"subjob ?\" \n";;
115	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
116	(*) printf "\n +++ unknown error"
117	printf "\n please inform S. Raasch!\n"
118	esac
119	[[ -f $job_to_send ]] && rm $job_to_send
120	printf "\n\n+++ SUBJOB killed \n\n"
121	fi' exit
122
123
124	# BEI TERMINAL-BREAK:
125	trap '[[ -f $job_to_send ]] && rm $job_to_send
126	printf "\n\n+++ SUBJOB killed \n\n"
127	exit
128	' 2
129
130
131
132
133	# LOKALEN HOSTNAMEN FESTSTELLEN
134	local_host=$(hostname)
135
136
137
138	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
139	# UEBERHAUPT ZULAESSIG IST
140	# Note: One of the entries for "lck" or "lckordi" always has to be
141	# comment out, because the hostname (node*) is same for both machines
142	case $local_host in
143	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
144	(atmos) local_addres=172.20.25.35; local_host=lcide;;
145	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
146	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
147	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
148	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
149	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
150	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
151	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
152	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
153	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
154	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
155	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
156	(blogin\|bxc) local_addres=130.73.233.1; local_host=lccrayb;;
157	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
158	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
159	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
160	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
161	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
162	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
163	(cs*) local_addres=136.172.44.131; local_host=nech;;
164	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
165	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
166	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
167	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
168	(node*) local_addres=165.132.26.61 local_host=lck;;
169	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
170	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
171	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
172	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
173	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
174	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
175	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
176	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
177	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
178	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
179	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
180	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
181	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
182	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
183	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
184	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
185	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
186	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
187	(jaboticaba) local_addres=150.163.25.181; local_host=lcbr;;
188	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
189	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
190	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
191	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
192	(meller) local_addres=134.106.74.155; local_host=lcfor;;
193	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
194	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
195	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
196	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
197	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
198	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
199	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
200	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
201	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
202	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
203	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
204	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
205	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
206	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
207	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
208	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
209	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
210	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
211	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
212	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
213	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
214	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
215	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
216	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
217	(*) printf "\n +++ \"$local_host\" unknown";
218	printf "\n please contact the PALM group at IMUK";
219	locat=parameter; exit;;
220	esac
221
222
223
224	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
225	remote_host=$local_host
226
227
228
229
230	# PROZEDUROPTIONEN EINLESEN
231	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
232	do
233	case $option in
234	(c) job_catalog=$OPTARG;;
235	(d) delete_dayfile=true;;
236	(D) no_submit=true;;
237	(e) email_notification=$OPTARG;;
238	(g) group_number=$OPTARG;;
239	(h) remote_host=$OPTARG;;
240	(m) memory=$OPTARG;;
241	(n) job_name=$OPTARG;;
242	(N) node_usage=$OPTARG;;
243	(O) threads_per_task=$OPTARG;;
244	(P) scp_port=$OPTARG;;
245	(q) no_default_queue=$OPTARG;;
246	(t) cputime=$OPTARG;;
247	(T) tasks_per_node=$OPTARG;;
248	(u) remote_user=$OPTARG;;
249	(v) verify=false;;
250	(X) numprocs=$OPTARG;;
251	(\?) printf "\n +++ Option $OPTARG unknown \n";
252	locat=option; exit;;
253	esac
254	done
255
256
257	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
258	(( to_shift = $OPTIND - 1 ))
259	shift $to_shift; file_to_send=$1
260
261
262	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
263	if [ "$1" = "?" ]
264	then
265	(printf "\n *** subjob can be called as follows:\n"
266	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
267	printf "\n Description of available options:\n"
268	printf "\n Option Description Default-Value"
269	printf "\n -c job-input- and output-catalog ~/job_queue"
270	printf "\n -d no job-protocol will be created ---"
271	printf "\n -D only the job-file will be created ---"
272	printf "\n -h execution host, available hosts: $remote_host"
273	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
274	printf "\n lckiaps, lctit, nech, necriam, unics"
275	printf "\n -m memory demand per process in MByte ---"
276	printf "\n -n jobname <jobdatei>"
277	printf "\n -O threads per task (for OpenMP usage) 1"
278	printf "\n -P ssh/scp port default port"
279	printf "\n -q job-queue to be used default"
280	printf "\n -t allowed cpu-time in seconds ---"
281	printf "\n -T tasks per node (on parallel hosts) ---"
282	printf "\n -u username on execution host from .netrc"
283	printf "\n -v no prompt for confirmation ---"
284	printf "\n -X # of processors (on parallel hosts) 1"
285	printf "\n "
286	printf "\n The only possible positional parameter is <jobfile>:"
287	printf "\n The complete NQS-job must be provided here."
288	printf "\n <jobfile>=? creates this outline\n\n") \| more
289	exit
290	fi
291
292
293
294	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
295	if [[ "$file_to_send" = "" ]]
296	then
297	printf "\n +++ job-file missing"
298	locat=parameter; exit
299	else
300	if [[ -f $file_to_send ]]
301	then
302	true
303	else
304	printf "\n +++ job-file: "
305	printf "\n $file_to_send"
306	printf "\n does not exist"
307	locat=parameter; exit
308	fi
309	fi
310
311
312
313	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
314	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
315	if [[ $job_name = none ]]
316	then
317	job_name=$file_to_send
318	fi
319	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
320	then
321	printf "\n +++ job-file name: "
322	printf "\n $job_name"
323	printf "\n must not contain \"/\"-characters"
324	locat=parameter; exit
325	fi
326
327
328
329
330	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
331	# ODER WENN HOST NICHT ANGEGEBEN WURDE
332	if [[ $remote_host = none ]]
333	then
334	printf "\n +++ host missing"
335	locat=option; exit
336	else
337	case $remote_host in
338	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
339	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
340	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
341	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
342	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
343	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
344	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/moab/7.2.6/bin/msub;;
345	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
346	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
347	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
348	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
349	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
350	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
351	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
352	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
353	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
354	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
355	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
356	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
357	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
358	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
359	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
360	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
361	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
362	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
363	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
364	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
365	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
366	locat=parameter; exit;;
367	esac
368	fi
369
370
371	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
372	if [[ $no_default_queue != none ]]
373	then
374	error=false
375	ndq=$no_default_queue
376	case $remote_host in
377	(ibm) case $ndq in
378	(p690_express\|p690_standard\|p690_long) error=false;;
379	(*) error=true;;
380	esac;;
381	(ibmh) case $ndq in
382	(cluster\|express) error=false;;
383	(*) error=true;;
384	esac;;
385	(ibmkisti) case $ndq in
386	(class.32plus\|class.1-2\|class.2-32) error=false;;
387	(*) error=true;;
388	esac;;
389	(ibmku) case $ndq in
390	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
391	(*) error=true;;
392	esac;;
393	(ibms) case $ndq in
394	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
395	(*) error=true;;
396	esac;;
397	(ibmy) case $ndq in
398	(parallel) error=false;;
399	(*) error=true;;
400	esac;;
401	(lccrayb) case $ndq in
402	(mpp1q\|mpp1testq\|testq) error=false;;
403	(*) error=true;;
404	esac;;
405	(lccrayf) case $ndq in
406	(usup\|test*\|small\|large) error=false;;
407	(*) error=true;;
408	esac;;
409	(lcflow) case $ndq in
410	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
411	(*) error=true;;
412	esac;;
413	(lckiaps) case $ndq in
414	(express\|normal) error=false;;
415	(*) error=true;;
416	esac;;
417	(lckyoto) case $ndq in
418	(eh\|ph) error=false;;
419	(*) error=true;;
420	esac;;
421	(lckyuh) case $ndq in
422	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
423	(*) error=true;;
424	esac;;
425	(lckyut) case $ndq in
426	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
427	(*) error=true;;
428	esac;;
429	(lcsgib\|lcsgih) case $ndq in
430	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
431	(*) error=true;;
432	esac;;
433	(lctit) case $ndq in
434	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
435	(*) error=true;;
436	esac;;
437	(t3eb) case $ndq in
438	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
439	(*) error=true;;
440	esac;;
441	(necriam) case $ndq in
442	(SP\|SS\|P6) error=false;;
443	(*) error=true;;
444	esac;;
445	(t3eh) case $ndq in
446	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
447	(*) error=true;;
448	esac;;
449	(t3ej2\|t3ej5) case $ndq in
450	(low\|normal\|high) error=false;;
451	(*) error=true;;
452	esac;;
453	(t3es) case $ndq in
454	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
455	(*) error=true;;
456	esac;;
457	(unics) case $ndq in
458	(unics\|ht) error=false;;
459	(*) error=true;;
460	esac;;
461	esac
462	if [[ $error = true ]]
463	then
464	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
465	locat=parameter; exit
466	else
467	queue=$no_default_queue
468	fi
469	fi
470
471
472
473	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
474	# AUFTEILEN
475	done=false
476	while [[ $done = false ]]
477	do
478	if (( $cputime <= 0 ))
479	then
480	printf "\n +++ wrong cpu-time or cpu-time missing"
481	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
482	printf "\n >>> "
483	read cputime 1>/dev/null 2>&1
484	else
485	done=true
486	fi
487	done
488	if [[ $remote_host = nech ]]
489	then
490	if (( tasks_per_node != 0 ))
491	then
492	(( cputime = cputime * tasks_per_node ))
493	elif [[ $numprocs != 0 ]]
494	then
495	(( cputime = cputime * numprocs ))
496	fi
497	fi
498	(( stunden = cputime / 3600 ))
499	(( resttime = cputime - stunden * 3600 ))
500	(( minuten = resttime / 60 ))
501	(( sekunden = resttime - minuten * 60 ))
502	timestring=${stunden}:${minuten}:${sekunden}
503
504
505
506	# PRUEFEN DER KERNSPEICHERANFORDERUNG
507	done=false
508	while [[ $done = false ]]
509	do
510	if (( memory <= 0 ))
511	then
512	printf "\n +++ wrong memory demand or memory demand missing"
513	printf "\n >>> Please type memory in MByte per process as INTEGER:"
514	printf "\n >>> "
515	read memory 1>/dev/null 2>&1
516	else
517	done=true
518	fi
519	done
520
521	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
522	then
523	if (( tasks_per_node != 0 ))
524	then
525	(( Memory = memory * tasks_per_node / 1000 ))
526	elif [[ $numprocs != 0 ]]
527	then
528	(( Memory = memory * numprocs / 1000 ))
529	else
530	(( Memory = memory / 1000 ))
531	fi
532	elif [[ $remote_host = lctit ]]
533	then
534	(( Memory = memory * tasks_per_node / 1000 ))
535	fi
536
537
538	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
539	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
540	then
541	(( memory = memory * threads_per_task ))
542	fi
543
544
545	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
546	if (( tasks_per_node != 0 ))
547	then
548	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
549	fi
550
551
552	# Calculate number of processes per node
553	(( processes_per_node = tasks_per_node * threads_per_task ))
554
555
556	# Calculate number of MPI tasks
557	(( mpi_tasks = numprocs / threads_per_task ))
558
559
560	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
561	if [[ "$scp_port" != "" ]]
562	then
563	PORTOPT="-P $scp_port"
564	SSH_PORTOPT="-p $scp_port"
565	fi
566
567
568	# HEADER-AUSGABE
569	if [[ $verify = true ]]
570	then
571	printf "\n\n"
572	printf "#--------------------------------------------------------------# \n"
573	spalte1=SUBJOB;spalte2=$(date)
574	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
575	printf "\| \| \n"
576	printf "\| values of parameters/options: \| \n"
577	spalte1=$(echo local_host$punkte \| cut -c-20)
578	spalte2=$punkte$local_host
579	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
580	spalte1=$(echo remote_host$punkte \| cut -c-20)
581	spalte2=$punkte$remote_host
582	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
583	spalte1=$(echo queue$punkte \| cut -c-20)
584	spalte2=$punkte$queue
585	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
586	spalte1=$(echo memory$punkte \| cut -c-20)
587	spalte2="$punkte$memory mb"
588	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
589	spalte1=$(echo cputime$punkte \| cut -c-20)
590	spalte2="$punkte$cputime sec"
591	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
592	spalte1=$(echo job_name$punkte \| cut -c-20)
593	spalte2="$punkte$job_name"
594	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
595	printf "#--------------------------------------------------------------# \n\n"
596
597
598	# KONTROLLABFRAGE, OB ALLES O.K.
599	antwort="dummy"
600	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
601	do
602	read antwort?" >>> continue (y/n) ? "
603	done
604	if [[ $antwort = n \|\| $antwort = N ]]
605	then
606	locat=verify; exit
607	fi
608	printf "\n"
609	fi
610
611
612
613	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
614	kennung=$RANDOM
615	job_on_remhost=${job_name}_${kennung}_$local_host
616	job_to_send=job_to_send_$kennung
617	if [[ $delete_dayfile = false ]]
618	then
619	remote_dayfile=${local_host}_${job_name}_result_$kennung
620	local_dayfile=${remote_host}_${job_name}
621	else
622	remote_dayfile=/dev/null
623	fi
624
625
626
627	# Generate the batch job scripts (qsub/msub/LoadLeveler)
628	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
629	then
630
631	# General LoadLeveler settings
632	execute_in_shell="#!/bin/ksh"
633	use_shell="# @ shell = /bin/ksh"
634	consumable_memory="ConsumableMemory($memory mb)"
635	class="# @ class = $queue"
636	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
637	network_to_use="# @ network.mpi = sn_all,shared,us"
638	data_limit="# @ data_limit = 1.76gb"
639	image_size="# @ image_size = 50"
640	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
641
642	if [[ $email_notification = none ]]
643	then
644	notify_user=""
645	else
646	notify_user="# @ notify_user = $email_notification"
647	if [[ $delete_dayfile = true ]]
648	then
649	notification='# @ notification = never'
650	fi
651	fi
652
653	if [[ $remote_host = ibmh ]]
654	then
655	data_limit=""
656	network_to_use=""
657	class="# @ class = $queue"
658	environment=""
659	rset="# @ rset = RSET_MCM_AFFINITY"
660	task_affinity="# @ task_affinity = core(1)"
661	elif [[ $remote_host = ibmkisti ]]
662	then
663	network_to_use="# @ network.MPI = sn_all,shared,US"
664	wall_clock_limit="# @ wall_clock_limit = $timestring"
665	if [[ $threads_per_task = 1 ]]
666	then
667	rset="# @ rset = RSET_MCM_AFFINITY"
668	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
669	fi
670	environment=""
671	use_shell=""
672	data_limit=""
673	image_size=""
674	elif [[ $remote_host = ibmku ]]
675	then
676	execute_in_shell="#!/usr/bin/ksh"
677	use_shell="# @ shell = /usr/bin/ksh"
678	consumable_memory=""
679	environment=""
680	network_to_use="# @ network.mpi = sn_all,shared,us"
681	data_limit=""
682	image_size=""
683	elif [[ $remote_host = ibms ]]
684	then
685	network_to_use="# @ network.mpi = csss,shared,us"
686	elif [[ $remote_host = ibmy ]]
687	then
688	consumable_memory=""
689	network_to_use=""
690	fi
691
692	cat > $job_to_send << %%END%%
693	$execute_in_shell
694	$use_shell
695
696	# @ job_type = parallel
697	# @ job_name = $job_name
698	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
699	# @ output = $remote_dayfile
700	# @ error = $remote_dayfile
701	$wall_clock_limit
702	$image_size
703	$class
704	$environment
705	$network_to_use
706	$data_limit
707	$rset
708	$mcm_affinity_options
709	$task_affinity
710	$notification
711	$notify_user
712
713	%%END%%
714
715	if (( nodes > 0 ))
716	then
717
718	if [[ $remote_host != ibmkisti ]]
719	then
720
721	cat >> $job_to_send << %%END%%
722	# @ node = $nodes
723	# @ tasks_per_node = $processes_per_node
724	# @ node_usage = $node_usage
725	# @ queue
726
727	%%END%%
728
729	else
730
731	cat >> $job_to_send << %%END%%
732	# @ total_tasks = $mpi_tasks
733	# @ blocking = unlimited
734	# @ queue
735
736	%%END%%
737
738	fi
739
740	else
741
742	if [[ $remote_host != ibmy ]]
743	then
744
745	cat >> $job_to_send << %%END%%
746	# @ blocking = unlimited
747	# @ total_tasks = $numprocs
748	# @ node_usage = $node_usage
749	# @ queue
750
751	%%END%%
752
753	else
754
755	cat >> $job_to_send << %%END%%
756	# @ node = 1
757	# @ total_tasks = $numprocs
758	# @ queue
759
760	%%END%%
761
762	fi
763
764	fi
765
766	# workaround because of silly job filter on ibmkisti
767	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
768	then
769	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
770	fi
771
772	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
773	then
774
775	cat > $job_to_send << %%END%%
776	#!/bin/ksh
777
778	# @ job_type = serial
779	# @ node_usage = $node_usage
780	# @ job_name = palm
781	# @ wall_clock_limit = ${timestring},$timestring
782	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
783	# @ output = $remote_dayfile
784	# @ error = $remote_dayfile
785	$class
786	$notification
787
788	# @ queue
789
790	%%END%%
791
792	elif [[ $remote_host = lccrayb ]]
793	then
794
795	if [[ $numprocs != 0 ]]
796	then
797	cat > $job_to_send << %%END%%
798	#!/bin/bash
799	#PBS -N $job_name
800	#PBS -l walltime=$timestring
801	#PBS -l nodes=$nodes:ppn=$processes_per_node
802	#PBS -o $remote_dayfile
803	#PBS -j oe
804	#PBS -q $queue
805
806	$module_calls
807
808	%%END%%
809
810	else
811
812	continue
813
814	fi
815
816	elif [[ $remote_host = lccrayf ]]
817	then
818
819	if [[ $numprocs != 0 ]]
820	then
821	cat > $job_to_send << %%END%%
822	#!/bin/bash -l
823	#SBATCH -J $job_name
824	#SBATCH -t $timestring
825	#SBATCH -N $nodes
826	#SBATCH --ntasks-per-node=$processes_per_node
827	#SBATCH -p $queue
828	#SBATCH -o $remote_dayfile
829	#SBATCH -e $remote_dayfile
830
831	$init_cmds
832	$module_calls
833
834	%%END%%
835
836	else
837	cat > $job_to_send << %%END%%
838	#!/bin/bash -l
839	#SBATCH -J $job_name
840	#SBATCH -t $timestring
841	#SBATCH -l ncpus=1
842	#SBATCH -l pmem=${memory}mb
843	#SBATCH -m abe
844	#SBATCH -o $remote_dayfile
845	#SBATCH -e $remote_dayfile
846
847	$init_cmds
848	$module_calls
849
850	%%END%%
851
852	fi
853
854	elif [[ $remote_host = lcfimm ]]
855	then
856
857	if [[ $numprocs != 0 ]]
858	then
859	cat > $job_to_send << %%END%%
860	#!/bin/ksh
861	#PBS -N $job_name
862	#PBS -A $project_account
863	#PBS -l walltime=$timestring
864	#PBS -l nodes=${nodes}:ppn=$processes_per_node
865	#PBS -l pmem=${memory}mb
866	#PBS -m abe
867	#PBS -o $remote_dayfile
868	#PBS -j oe
869	mpd &
870
871	%%END%%
872
873	else
874	cat > $job_to_send << %%END%%
875	#!/bin/ksh
876	#PBS -N $job_name
877	#PBS -A $project_account
878	#PBS -l walltime=$timestring
879	#PBS -l ncpus=1
880	#PBS -l pmem=${memory}mb
881	#PBS -m abe
882	#PBS -o $remote_dayfile
883	#PBS -j oe
884
885	%%END%%
886
887	fi
888
889	elif [[ $remote_host = lcflow ]]
890	then
891	if [ $memory -gt 1800 ]; then
892	use_himem=""
893	else
894	use_himem="#"
895	fi
896	if [[ $numprocs != 0 ]]
897	then
898	pe_set="#$ -pe impi41 $numprocs"
899	else
900	pe_set="#$ -pe impi41 1"
901	fi
902	if [[ $queue = default ]]
903	then
904	queue_set=""
905	else
906	queue_set="#$ -q $queue"
907	fi
908	[[ "$disc_space" = "" ]] && disc_space=50
909
910	cat > $job_to_send << %%END%%
911	#!/bin/bash
912	#$ -S /bin/bash
913	#$ -N $job_name
914	#$ -cwd
915	#$ -l h_rt=$timestring
916	#$ -l h_vmem=${memory}M
917	#$ -o $remote_dayfile
918	#$ -j y
919	$pe_set
920	#$ -R y
921	${use_himem}#$ -l highmem=true
922	#$ -l h_fsize=${disc_space}G
923	$queue_set
924
925	%%END%%
926
927	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
928	then
929
930	if [[ $numprocs != 0 ]]
931	then
932	cat > $job_to_send << %%END%%
933	#!/bin/ksh
934	#PBS -N $job_name
935	#PBS -l walltime=$timestring
936	#PBS -l ncpus=$numprocs
937	#PBS -l pmem=${memory}mb
938	#PBS -o $remote_dayfile
939	#PBS -l nodes=$nodes:ppn=${processes_per_node}
940	#PBS -j oe
941
942	mpd &
943
944	%%END%%
945
946	else
947	cat > $job_to_send << %%END%%
948	#!/bin/ksh
949	#PBS -N $job_name
950	#PBS -l walltime=$timestring
951	#PBS -l ncpus=1
952	#PBS -l pmem=${memory}mb
953	#PBS -o $remote_dayfile
954	#PBS -j oe
955
956	%%END%%
957
958	fi
959
960	elif [[ $remote_host = lckiaps ]]
961	then
962
963	if [[ $numprocs != 0 ]]
964	then
965	cat > $job_to_send << %%END%%
966	#!/bin/ksh
967	#PBS -N $job_name
968	#PBS -l walltime=$timestring
969	#PBS -l select=1:ncpus=$numprocs
970	#PBS -l pmem=${memory}mb
971	#PBS -q $queue
972	#PBS -o $remote_dayfile
973	#PBS -j oe
974	#PBS -V
975
976	%%END%%
977
978	else
979	cat > $job_to_send << %%END%%
980	#!/bin/ksh
981	#PBS -N $job_name
982	#PBS -l walltime=$timestring
983	#PBS -l ncpus=1
984	#PBS -l pmem=${memory}mb
985	#PBS -o $remote_dayfile
986	#PBS -j oe
987
988	%%END%%
989
990	fi
991
992	elif [[ $remote_host = lcyon ]]
993	then
994
995	if [[ $numprocs != 0 ]]
996	then
997	cat > $job_to_send << %%END%%
998	#!/bin/ksh
999	#PBS -N $job_name
1000	#PBS -l walltime=$timestring
1001	#PBS -l ncpus=$numprocs
1002	#PBS -l pmem=${memory}mb
1003	#PBS -o $remote_dayfile
1004	#PBS -j oe
1005
1006	%%END%%
1007
1008	else
1009	cat > $job_to_send << %%END%%
1010	#!/bin/ksh
1011	#PBS -N $job_name
1012	#PBS -l walltime=$timestring
1013	#PBS -l ncpus=1
1014	#PBS -l pmem=${memory}mb
1015	#PBS -o $remote_dayfile
1016	#PBS -j oe
1017
1018	%%END%%
1019
1020	fi
1021
1022	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1023	then
1024
1025	if [[ "$sgi_feature" != "" ]]
1026	then
1027	feature_directive="#PBS -l feature=$sgi_feature"
1028	else
1029	feature_directive=""
1030	fi
1031
1032	if [[ $queue = dataq \|\| $queue = permq ]]
1033	then
1034	feature_directive="#PBS -l feature=data"
1035	fi
1036
1037	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
1038	then
1039	queue_directive="#PBS -q $queue"
1040	else
1041	queue_directive=""
1042	fi
1043
1044	if [[ $email_notification = none ]]
1045	then
1046	email_directive=""
1047	else
1048	email_directive="#PBS -M $email_notification"
1049	fi
1050
1051	if [[ $numprocs != 0 ]]
1052	then
1053	cat > $job_to_send << %%END%%
1054	#!/bin/bash
1055	#PBS -N $job_name
1056	#PBS -l walltime=$timestring
1057	#PBS -l nodes=$nodes:ppn=${processes_per_node}
1058	#PBS -l naccesspolicy=$node_usage
1059	#PBS -o $remote_dayfile
1060	#PBS -j oe
1061	$feature_directive
1062	$queue_directive
1063	$email_directive
1064
1065	eval \`/sw/swdist/bin/modulesinit\`
1066	#. /usr/share/modules/init/bash
1067	$init_cmds
1068	$module_calls
1069
1070	echo ld_library_path=\$LD_LIBRARY_PATH
1071
1072	%%END%%
1073
1074	else
1075	cat > $job_to_send << %%END%%
1076	#PBS -S /bin/bash
1077	#PBS -N $job_name
1078	#PBS -l walltime=$timestring
1079	#PBS -l ncpus=1
1080	#PBS -o $remote_dayfile
1081	#PBS -j oe
1082	$queue_directive
1083	$email_directive
1084
1085	eval \`/sw/swdist/bin/modulesinit\`
1086	#. /usr/share/modules/init/bash
1087	$init_cmds
1088	$module_calls
1089
1090	%%END%%
1091
1092	fi
1093
1094	elif [[ $remote_host = lcxe6 ]]
1095	then
1096
1097	if [[ $numprocs != 0 ]]
1098	then
1099	cat > $job_to_send << %%END%%
1100	#!/bin/ksh
1101	#PBS -S /bin/ksh
1102	#PBS -N $job_name
1103	#PBS -A $project_account
1104	#PBS -j oe
1105	#PBS -l walltime=$timestring
1106	#PBS -l mppwidth=${numprocs}
1107	#PBS -l mppnppn=${processes_per_node}
1108	#PBS -m abe
1109	#PBS -o $remote_dayfile
1110	$email_directive
1111
1112	$init_cmds
1113	$module_calls
1114
1115	%%END%%
1116
1117	else
1118	cat > $job_to_send << %%END%%
1119	#!/bin/ksh
1120	#PBS -S /bin/ksh
1121	#PBS -N $job_name
1122	#PBS -A $project_account
1123	#PBS -j oe
1124	#PBS -l walltime=$timestring
1125	#PBS -l ncpus=1
1126	#PBS -l pmem=${memory}mb
1127	#PBS -m abe
1128	$email_directive
1129	#PBS -o $remote_dayfile
1130
1131	$init_cmds
1132	$module_calls
1133
1134	%%END%%
1135
1136	fi
1137
1138	elif [[ $remote_host = lckyoto ]]
1139	then
1140
1141	cat > $job_to_send << %%END%%
1142	#!/bin/ksh
1143	# @\$-o $remote_dayfile
1144	# @\$-eo -oi
1145	# @\$-lP 16
1146	# @\$-lp 1
1147	# @\$-lm 28gb -llm unlimited -ls unlimited
1148	# @\$-q $queue
1149	# @\$-Pvn abs_pack
1150	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1151	#. /thin/local/etc/setprofile/intel-11.0.sh
1152	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1153	. ~/.myprofile
1154	#. /home2/t/t51254/palm/current_version/myprofile
1155	#. /thin/apps/pgi/mpi.sh
1156	#
1157	env
1158	#
1159	set -x
1160
1161	%%END%%
1162
1163	elif [[ $remote_host = lcxt5m ]]
1164	then
1165
1166	if [[ $numprocs != 0 ]]
1167	then
1168	cat > $job_to_send << %%END%%
1169	#!/bin/ksh
1170	#PBS -S /bin/ksh
1171	#PBS -N $job_name
1172	#PBS -j oe
1173	#PBS -l walltime=$timestring
1174	#PBS -l mppwidth=${numprocs}
1175	#PBS -l mppnppn=${processes_per_node}
1176	#PBS -m abe
1177	#PBS -o $remote_dayfile
1178
1179	$init_cmds
1180	$module_calls
1181
1182	%%END%%
1183
1184	else
1185	cat > $job_to_send << %%END%%
1186	#!/bin/ksh
1187	#PBS -S /bin/ksh
1188	#PBS -N $job_name
1189	#PBS -j oe
1190	#PBS -l walltime=$timestring
1191	#PBS -l ncpus=1
1192	#PBS -l pmem=${memory}mb
1193	#PBS -m abe
1194	#PBS -o $remote_dayfile
1195
1196	$init_cmds
1197	$module_calls
1198
1199	%%END%%
1200
1201	fi
1202
1203	elif [[ $remote_host = lckyuh ]]
1204	then
1205	cat > $job_to_send << %%END%%
1206	#!/bin/bash
1207	#PJM -L "rscgrp=$queue"
1208	#PJM -L "node=$nodes"
1209	#PJM --mpi "proc=$numprocs"
1210	#PJM -L "elapse=$timestring"
1211	#PJM -o $remote_dayfile
1212	#PJM -j
1213	#PJM -X
1214	#PJM --no-stging
1215
1216	export LANG=en_US.UTF-8
1217	%%END%%
1218
1219	elif [[ $remote_host = lckyut ]]
1220	then
1221	cat > $job_to_send << %%END%%
1222	#!/bin/bash
1223	#PJM -L "rscgrp=$queue"
1224	#PJM -L "vnode=$numprocs"
1225	#PJM -L "vnode-core=1"
1226	#PJM -L "elapse=$timestring"
1227	#PJM --mpi proc=$numprocs
1228	#PJM -o $remote_dayfile
1229	#PJM -j
1230	#PJM -X
1231	#PJM --no-stging
1232
1233	export LANG=en_US.UTF-8
1234	%%END%%
1235
1236	elif [[ $remote_host = nech ]]
1237	then
1238
1239	if (( nodes > 1 ))
1240	then
1241	cat > $job_to_send << %%END%%
1242	#!/bin/ksh
1243	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1244	#PBS -l ${qsubmem}=${Memory}gb
1245	#PBS -b $nodes
1246	#PBS -o $remote_dayfile
1247	#PBS -N palm
1248	#PBS -j o
1249	#PBS -T mpisx
1250
1251	%%END%%
1252
1253	elif [[ $numprocs != 0 ]]
1254	then
1255	cat > $job_to_send << %%END%%
1256	#!/bin/ksh
1257	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1258	#PBS -l ${qsubmem}=${Memory}gb
1259	#PBS -o $remote_dayfile
1260	#PBS -N palm
1261	#PBS -j o
1262
1263	%%END%%
1264
1265	else
1266	cat > $job_to_send << %%END%%
1267	#!/bin/ksh
1268	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1269	#PBS -o $remote_dayfile
1270	#PBS -j o
1271
1272	%%END%%
1273
1274	fi
1275
1276	elif [[ $remote_host = necriam ]]
1277	then
1278
1279	if (( nodes > 1 ))
1280	then
1281	cat > $job_to_send << %%END%%
1282	#!/bin/ksh
1283	#PBS -b $nodes
1284	#PBS -o $remote_dayfile
1285	#PBS -N $job_name
1286	#PBS -j o
1287	#PBS -v MPIPROGINV=YES
1288
1289	%%END%%
1290
1291	elif [[ $numprocs != 0 ]]
1292	then
1293	cat > $job_to_send << %%END%%
1294	#!/bin/ksh
1295	#PBS -q ${queue}
1296	#PBS -o $remote_dayfile
1297	#PBS -N $job_name
1298	#PBS -j o
1299	#PBS -v MPIPROGINV=YES
1300
1301	%%END%%
1302
1303	else
1304	cat > $job_to_send << %%END%%
1305	#!/bin/ksh
1306	#PBS -o $remote_dayfile
1307	#PBS -j o
1308
1309	%%END%%
1310
1311	fi
1312
1313	elif [[ $remote_host = lctit ]]
1314	then
1315	cat > $job_to_send << %%END%%
1316	#!/bin/ksh
1317	$init_cmds
1318	$module_calls
1319
1320	%%END%%
1321
1322	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1323	if [[ $tasks_per_node != $processes_per_node ]]
1324	then
1325	submcom="$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 $queue "
1326	else
1327	submcom="$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 "
1328	fi
1329
1330	else
1331
1332	cat > $job_to_send << %%END%%
1333	# @\$-q ${queue}
1334	# @\$-l${qsubtime} $timestring
1335	# @\$-l${qsubmem} ${memory}mb
1336	# @\$-o $remote_dayfile
1337	# @\$-eo
1338
1339	%%END%%
1340
1341	fi
1342
1343
1344	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1345	# BEI EXIT VERANLASSEN
1346	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1347	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1348	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1349	then
1350	echo "set +vx" >> $job_to_send
1351	echo "trap '" >> $job_to_send
1352	echo "set +vx" >> $job_to_send
1353	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* ]]
1354	then
1355	if [[ $remote_host = ibmh ]]
1356	then
1357	return_queue=c1
1358	elif [[ $remote_host = ibmkisti ]]
1359	then
1360	return_queue=class.1-2
1361	elif [[ $remote_host = ibmku ]]
1362	then
1363	return_queue=sdbg2
1364	elif [[ $remote_host = ibms ]]
1365	then
1366	return_queue=p_normal
1367	elif [[ $remote_host = ibmy ]]
1368	then
1369	return_queue=serial
1370	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1371	then
1372	return_queue=serialq
1373	elif [[ $remote_host = necriam ]]
1374	then
1375	return_queue=SP
1376	elif [[ $remote_host = lccrayb ]]
1377	then
1378	return_queue=dataq
1379	elif [[ $remote_host = lckiaps ]]
1380	then
1381	return_queue=express
1382	elif [[ $remote_host = lckyuh ]]
1383	then
1384	return_queue=cx-single
1385	elif [[ $remote_host = lckyut ]]
1386	then
1387	return_queue=cx-single
1388	else
1389	return_queue=unknown
1390	fi
1391
1392	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1393	then
1394
1395	if [[ $remote_host = ibmku ]]
1396	then
1397	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1398	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1399	else
1400	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1401	fi
1402	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1403	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1404	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1405	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1406	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1407	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1408	if [[ $host != "ibmh" ]]
1409	then
1410	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1411	fi
1412	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1413	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1414
1415	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1416	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1417
1418	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1419	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
1420	if [[ $remote_host = ibmku ]]
1421	then
1422	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1423	fi
1424	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1425
1426	elif [[ $remote_host = nech ]]
1427	then
1428	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1429	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1430	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1431	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1432	echo "#PBS -j o" >> $job_to_send
1433	echo " " >> $job_to_send
1434	echo "set -x" >> $job_to_send
1435	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1436	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1437	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1438	echo "%%END%%" >> $job_to_send
1439
1440	elif [[ $remote_host = necriam ]]
1441	then
1442	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1443	echo "#PBS -q $return_queue" >> $job_to_send
1444	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1445	echo "#PBS -j o" >> $job_to_send
1446	echo " " >> $job_to_send
1447	echo "set -x" >> $job_to_send
1448	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1449	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1450	echo "%%END%%" >> $job_to_send
1451
1452	elif [[ $remote_host = lckyuh ]]
1453	then
1454	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1455	echo "#!/bin/bash" >> $job_to_send
1456	echo "#PJM -L \"node=1\"" >> $job_to_send
1457	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1458	echo "#PJM --no-stging" >> $job_to_send
1459	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1460	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1461	echo "#PJM -j" >> $job_to_send
1462	echo " " >> $job_to_send
1463	echo "export LANG=en_US.UTF-8" >> $job_to_send
1464	echo "set -x" >> $job_to_send
1465	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1466	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1467	echo "%%END%%" >> $job_to_send
1468
1469	elif [[ $remote_host = lckyut ]]
1470	then
1471	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1472	echo "#!/bin/bash" >> $job_to_send
1473	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1474	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1475	echo "#PJM --no-stging" >> $job_to_send
1476	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1477	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1478	echo "#PJM -j" >> $job_to_send
1479	echo " " >> $job_to_send
1480	echo "export LANG=en_US.UTF-8" >> $job_to_send
1481	echo "set -x" >> $job_to_send
1482	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1483	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1484	echo "%%END%%" >> $job_to_send
1485
1486	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1487	then
1488	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1489	echo "#!/bin/bash" >> $job_to_send
1490	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1491	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1492	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1493	echo "#PBS -l feature=data" >> $job_to_send
1494	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1495	echo "#PBS -j oe" >> $job_to_send
1496	echo " " >> $job_to_send
1497	echo ". /usr/share/modules/init/bash" >> $job_to_send
1498	echo "set -x" >> $job_to_send
1499	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1500	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1501	echo "%%END%%" >> $job_to_send
1502
1503	elif [[ $remote_host = lccrayb ]]
1504	then
1505	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1506	echo "#!/bin/bash" >> $job_to_send
1507	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1508	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1509	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1510	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1511	echo "#PBS -j oe" >> $job_to_send
1512	echo " " >> $job_to_send
1513	echo "set -x" >> $job_to_send
1514	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1515	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1516	echo "%%END%%" >> $job_to_send
1517
1518	elif [[ $remote_host = lcflow ]]
1519	then
1520	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1521	echo "#!/bin/bash" >> $job_to_send
1522	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1523	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1524	echo "SGEPREFIX -cwd" >> $job_to_send
1525	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1526	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1527	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1528	echo "SGEPREFIX -j y" >> $job_to_send
1529	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1530	echo " " >> $job_to_send
1531	echo "set -x" >> $job_to_send
1532	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1533	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1534	echo "" >> $job_to_send
1535	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1536	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1537	echo "rm -f scpjob.${kennung}" >> $job_to_send
1538	echo "%%END%%" >> $job_to_send
1539	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1540	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1541
1542	else
1543
1544	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1545	echo "# @\\\$-q $return_queue" >> $job_to_send
1546	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1547	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1548	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1549	then
1550	echo "# @\$-l mpp_p=0" >> $job_to_send
1551	fi
1552	echo '# @\$-lF 10mb' >> $job_to_send
1553	# echo '# @\$-o /dev/null' >> $job_to_send
1554	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1555	echo '# @\\\$-eo' >> $job_to_send
1556	echo " " >> $job_to_send
1557	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1558	then
1559	echo "set +vx" >> $job_to_send
1560	echo ". .profile" >> $job_to_send
1561	fi
1562	echo "set -x" >> $job_to_send
1563	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
1564	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1565	echo "%%END%%" >> $job_to_send
1566	fi
1567
1568	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1569	then
1570	echo "llsubmit scpjob.$kennung" >> $job_to_send
1571	elif [[ $remote_host = lccrayb ]]
1572	then
1573	echo "msub -q $return_queue scpjob.$kennung" >> $job_to_send
1574	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1575	then
1576	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1577	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1578	echo "msub scpjob.$kennung" >> $job_to_send
1579	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1580	then
1581	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1582	elif [[ $remote_host = t3es ]]
1583	then
1584	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1585	elif [[ $remote_host = lckiaps ]]
1586	then
1587	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1588	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1589	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1590	elif [[ $remote_host = lckyu* ]]
1591	then
1592	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1593	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1594	elif [[ $remote_host = lcflow ]]
1595	then
1596	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1597	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1598	else
1599	echo "$submcom scpjob.$kennung" >> $job_to_send
1600	fi
1601	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1602	then
1603	echo "rm scpjob.$kennung" >> $job_to_send
1604	fi
1605	if [[ $remote_host = nech ]]
1606	then
1607	echo "cd -" >> $job_to_send
1608	fi
1609	else
1610	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1611	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1612	fi
1613	echo "set -x" >> $job_to_send
1614	echo " ' exit" >> $job_to_send
1615	echo "set -x" >> $job_to_send
1616	fi
1617
1618
1619
1620	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1621	cat $file_to_send >> $job_to_send
1622
1623	if [[ $remote_host = ibm ]]
1624	then
1625	echo " " >> $job_to_send
1626	echo "exit" >> $job_to_send
1627	fi
1628
1629	# remove job file
1630	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1631	then
1632	echo " " >> $job_to_send
1633	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1634	fi
1635
1636
1637
1638	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1639	if [[ -z $remote_user ]]
1640	then
1641	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1642	then
1643	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1644	fi
1645	fi
1646
1647
1648
1649	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1650	if [[ $no_submit = false ]]
1651	then
1652	if [[ $remote_host != $local_host ]]
1653	then
1654	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1655	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1656	then
1657	job_catalog_save=$job_catalog
1658	job_catalog=job_queue
1659	elif [[ $remote_host = nech ]]
1660	then
1661	job_catalog_save=$job_catalog
1662	job_catalog=/hpf/b/${remote_user}/job_queue
1663	fi
1664	if [[ $remote_host = nech ]]
1665	then
1666	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1667	# TRANSFERIERT WERDEN
1668	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1669	else
1670	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1671	fi
1672	if [[ $? = 1 ]]
1673	then
1674	locat=scp; exit
1675	fi
1676	if [[ $remote_host = ibms ]]
1677	then
1678	job_catalog=$job_catalog_save
1679	fi
1680	[[ $verify = true ]] && printf "\n >>> finished\n"
1681	else
1682	eval job_catalog=$job_catalog
1683	cp $job_to_send ${job_catalog}/$job_on_remhost
1684	fi
1685
1686
1687
1688	# NQS- BZW. LOADLEVELER-JOB STARTEN
1689	if [[ $remote_host != $local_host ]]
1690	then
1691	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1692
1693	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1694	then
1695	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1696	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1697	elif [[ $remote_host = ibmku ]]
1698	then
1699	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1700	elif [[ $remote_host = lcflow ]]
1701	then
1702	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1703	else
1704	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1705	fi
1706
1707	[[ $verify = true ]] && printf " >>> o.k.\n"
1708	else
1709	cd $job_catalog
1710	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1711	then
1712	eval $submcom $job_on_remhost
1713	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1714	then
1715	chmod u+x $job_on_remhost
1716	eval $submcom $job_on_remhost
1717	elif [[ $local_host = nech ]]
1718	then
1719	if [[ $queue = default ]]
1720	then
1721	eval $submcom $job_on_remhost
1722	else
1723	eval $submcom -q $queue $job_on_remhost
1724	fi
1725	else
1726	qsub $job_on_remhost
1727	fi
1728
1729	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1730	# only at the end of the job.
1731	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1732	then
1733	rm $job_on_remhost
1734	fi
1735	cd - > /dev/null
1736	fi
1737	fi
1738
1739
1740
1741	# ABSCHLUSSARBEITEN
1742
1743	if [[ $no_submit = false ]]
1744	then
1745	rm -f $job_to_send
1746	fi
1747	[[ $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 |