Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1256

Last change on this file since 1256 was 1256, checked in by raasch, 10 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 58.1 KB

Line
1	#!/bin/ksh
2
3	# subjob - script for automatic generation and submission of batch-job files
4	# for various batch queuing systems
5
6	#--------------------------------------------------------------------------------#
7	# This file is part of PALM.
8	#
9	# PALM is free software: you can redistribute it and/or modify it under the terms
10	# of the GNU General Public License as published by the Free Software Foundation,
11	# either version 3 of the License, or (at your option) any later version.
12	#
13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
16	#
17	# You should have received a copy of the GNU General Public License along with
18	# PALM. If not, see <http://www.gnu.org/licenses/>.
19	#
20	# Copyright 1997-2012 Leibniz University Hannover
21	#--------------------------------------------------------------------------------#
22	#
23	# Current revisions:
24	# ------------------
25	#
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1256 2013-11-07 14:46:21Z raasch $
30	#
31	# 1255 2013-11-07 14:43:35Z raasch
32	# further adjustments for lccrayb
33	#
34	# 1224 2013-09-16 07:27:23Z raasch
35	# first adjustments for lccrayb
36	#
37	# 1202 2013-07-10 16:22:07Z witha
38	# adjustments for Forwind cluster (lcflow)
39	#
40	# 1199 2013-07-05 14:52:22Z raasch
41	# adjustments for CSC Helsinki (lccrayf)
42	#
43	# use of cluster/express queue enabled (ibmh)
44	# vinessa added (imuk)
45	#
46	# 1103 2013-02-20 02:15:53Z raasch
47	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
48	# instead typeset -L/R),
49	# further adjustments for lckyuh
50	#
51	# 2013-02-10 01:47:43Z raasch
52	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
53	# and for Forwind cluster (lcflow)
54	#
55	# 1094 2013-02-03 01:52:12Z raasch
56	# new option -P for explicit setting of ssh/scp port,
57	# decalpha parts (yonsei) removed
58	#
59	# 2013-02-02 07:06:13Z raasch
60	# adjustments for Kyushu-University computing center (lckyut - tatara)
61	# old changelog messages removed
62	#
63	# 1046 2012-11-09 14:38:45Z maronga
64	# code put under GPL (PALM 3.9)
65	#
66	# 08/07/94 - Siggi - first version finished
67	# 29/06/94 - Siggi - script development started
68	#--------------------------------------------------------------------------------#
69	# subjob - script for automatic generation and submission of batch-job files
70	# for various batch queuing systems
71	#--------------------------------------------------------------------------------#
72
73
74	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
75	delete_dayfile=false
76	email_notification=none
77	group_number=none
78	locat=normal
79	no_default_queue=none
80	no_submit=false
81	job_catalog="~/job_queue"
82	job_name=none
83	local_user=$LOGNAME
84	node_usage=shared
85	numprocs=0
86	punkte="..........................................................."
87	submcom=qsub
88	queue=default
89	remote_host=none
90	remote_user=""
91	verify=true
92
93	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
94	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
95
96
97
98	# FEHLERBEHANDLUNG
99	# BEI EXIT:
100	trap 'if [[ $locat != normal ]]
101	then
102	case $locat in
103	(option) printf "\n --> available optios can be displayed"
104	printf " by typing:"
105	printf "\n \"subjob ?\" \n";;
106	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
107	(*) printf "\n +++ unknown error"
108	printf "\n please inform S. Raasch!\n"
109	esac
110	[[ -f $job_to_send ]] && rm $job_to_send
111	printf "\n\n+++ SUBJOB killed \n\n"
112	fi' exit
113
114
115	# BEI TERMINAL-BREAK:
116	trap '[[ -f $job_to_send ]] && rm $job_to_send
117	printf "\n\n+++ SUBJOB killed \n\n"
118	exit
119	' 2
120
121
122
123
124	# LOKALEN HOSTNAMEN FESTSTELLEN
125	local_host=$(hostname)
126
127
128
129	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
130	# UEBERHAUPT ZULAESSIG IST
131	# Note: One of the entries for "lck" or "lckordi" always has to be
132	# comment out, because the hostname (node*) is same for both machines
133	case $local_host in
134	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
135	(atmos) local_addres=172.20.25.35; local_host=lcide;;
136	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
137	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
138	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
139	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
140	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
141	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
142	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
143	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
144	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
145	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
146	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
147	(blogin\|bxc) local_addres=130.73.233.1; local_host=lccrayb;;
148	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
149	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
150	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
151	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
152	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
153	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
154	(cs*) local_addres=136.172.44.131; local_host=nech;;
155	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
156	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
157	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
158	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
159	(node*) local_addres=165.132.26.61 local_host=lck;;
160	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
161	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
162	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
163	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
164	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
165	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
166	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
167	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
168	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
169	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
170	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
171	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
172	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
173	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
174	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
175	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
176	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
177	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
178	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
179	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
180	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
181	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
182	(meller) local_addres=134.106.74.155; local_host=lcfor;;
183	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
184	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
185	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
186	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
187	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
188	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
189	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
190	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
191	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
192	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
193	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
194	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
195	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
196	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
197	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
198	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
199	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
200	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
201	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
202	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
203	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
204	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
205	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
206	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
207	(*) printf "\n +++ \"$local_host\" unknown";
208	printf "\n please contact the PALM group at IMUK";
209	locat=parameter; exit;;
210	esac
211
212
213
214	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
215	remote_host=$local_host
216
217
218
219
220	# PROZEDUROPTIONEN EINLESEN
221	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
222	do
223	case $option in
224	(c) job_catalog=$OPTARG;;
225	(d) delete_dayfile=true;;
226	(D) no_submit=true;;
227	(e) email_notification=$OPTARG;;
228	(g) group_number=$OPTARG;;
229	(h) remote_host=$OPTARG;;
230	(m) memory=$OPTARG;;
231	(n) job_name=$OPTARG;;
232	(N) node_usage=$OPTARG;;
233	(O) threads_per_task=$OPTARG;;
234	(P) scp_port=$OPTARG;;
235	(q) no_default_queue=$OPTARG;;
236	(t) cputime=$OPTARG;;
237	(T) tasks_per_node=$OPTARG;;
238	(u) remote_user=$OPTARG;;
239	(v) verify=false;;
240	(X) numprocs=$OPTARG;;
241	(\?) printf "\n +++ Option $OPTARG unknown \n";
242	locat=option; exit;;
243	esac
244	done
245
246
247	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
248	(( to_shift = $OPTIND - 1 ))
249	shift $to_shift; file_to_send=$1
250
251
252	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
253	if [ "$1" = "?" ]
254	then
255	(printf "\n *** subjob can be called as follows:\n"
256	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
257	printf "\n Description of available options:\n"
258	printf "\n Option Description Default-Value"
259	printf "\n -c job-input- and output-catalog ~/job_queue"
260	printf "\n -d no job-protocol will be created ---"
261	printf "\n -D only the job-file will be created ---"
262	printf "\n -h execution host, available hosts: $remote_host"
263	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
264	printf "\n lckiaps, lctit, nech, necriam, unics"
265	printf "\n -m memory demand per process in MByte ---"
266	printf "\n -n jobname <jobdatei>"
267	printf "\n -O threads per task (for OpenMP usage) 1"
268	printf "\n -P ssh/scp port default port"
269	printf "\n -q job-queue to be used default"
270	printf "\n -t allowed cpu-time in seconds ---"
271	printf "\n -T tasks per node (on parallel hosts) ---"
272	printf "\n -u username on execution host from .netrc"
273	printf "\n -v no prompt for confirmation ---"
274	printf "\n -X # of processors (on parallel hosts) 1"
275	printf "\n "
276	printf "\n The only possible positional parameter is <jobfile>:"
277	printf "\n The complete NQS-job must be provided here."
278	printf "\n <jobfile>=? creates this outline\n\n") \| more
279	exit
280	fi
281
282
283
284	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
285	if [[ "$file_to_send" = "" ]]
286	then
287	printf "\n +++ job-file missing"
288	locat=parameter; exit
289	else
290	if [[ -f $file_to_send ]]
291	then
292	true
293	else
294	printf "\n +++ job-file: "
295	printf "\n $file_to_send"
296	printf "\n does not exist"
297	locat=parameter; exit
298	fi
299	fi
300
301
302
303	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
304	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
305	if [[ $job_name = none ]]
306	then
307	job_name=$file_to_send
308	fi
309	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
310	then
311	printf "\n +++ job-file name: "
312	printf "\n $job_name"
313	printf "\n must not contain \"/\"-characters"
314	locat=parameter; exit
315	fi
316
317
318
319
320	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
321	# ODER WENN HOST NICHT ANGEGEBEN WURDE
322	if [[ $remote_host = none ]]
323	then
324	printf "\n +++ host missing"
325	locat=option; exit
326	else
327	case $remote_host in
328	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
329	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
330	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
331	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
332	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
333	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
334	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/torque/4.2.4.1/bin/qsub;;
335	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
336	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
337	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
338	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
339	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
340	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
341	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
342	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
343	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
344	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
345	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
346	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
347	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
348	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
349	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
350	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
351	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
352	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
353	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
354	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
355	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
356	locat=parameter; exit;;
357	esac
358	fi
359
360
361	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
362	if [[ $no_default_queue != none ]]
363	then
364	error=false
365	ndq=$no_default_queue
366	case $remote_host in
367	(ibm) case $ndq in
368	(p690_express\|p690_standard\|p690_long) error=false;;
369	(*) error=true;;
370	esac;;
371	(ibmh) case $ndq in
372	(cluster\|express) error=false;;
373	(*) error=true;;
374	esac;;
375	(ibmkisti) case $ndq in
376	(class.32plus\|class.1-2\|class.2-32) error=false;;
377	(*) error=true;;
378	esac;;
379	(ibmku) case $ndq in
380	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
381	(*) error=true;;
382	esac;;
383	(ibms) case $ndq in
384	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
385	(*) error=true;;
386	esac;;
387	(ibmy) case $ndq in
388	(parallel) error=false;;
389	(*) error=true;;
390	esac;;
391	(lccrayb) case $ndq in
392	(mpp1q\|mpp1testq\|testq) error=false;;
393	(*) error=true;;
394	esac;;
395	(lccrayf) case $ndq in
396	(usup\|test*\|small\|large) error=false;;
397	(*) error=true;;
398	esac;;
399	(lcflow) case $ndq in
400	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
401	(*) error=true;;
402	esac;;
403	(lckiaps) case $ndq in
404	(express\|normal) error=false;;
405	(*) error=true;;
406	esac;;
407	(lckyoto) case $ndq in
408	(eh\|ph) error=false;;
409	(*) error=true;;
410	esac;;
411	(lckyuh) case $ndq in
412	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
413	(*) error=true;;
414	esac;;
415	(lckyut) case $ndq in
416	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
417	(*) error=true;;
418	esac;;
419	(lcsgib\|lcsgih) case $ndq in
420	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
421	(*) error=true;;
422	esac;;
423	(lctit) case $ndq in
424	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
425	(*) error=true;;
426	esac;;
427	(t3eb) case $ndq in
428	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
429	(*) error=true;;
430	esac;;
431	(necriam) case $ndq in
432	(SP\|SS\|P6) error=false;;
433	(*) error=true;;
434	esac;;
435	(t3eh) case $ndq in
436	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
437	(*) error=true;;
438	esac;;
439	(t3ej2\|t3ej5) case $ndq in
440	(low\|normal\|high) error=false;;
441	(*) error=true;;
442	esac;;
443	(t3es) case $ndq in
444	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
445	(*) error=true;;
446	esac;;
447	(unics) case $ndq in
448	(unics\|ht) error=false;;
449	(*) error=true;;
450	esac;;
451	esac
452	if [[ $error = true ]]
453	then
454	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
455	locat=parameter; exit
456	else
457	queue=$no_default_queue
458	fi
459	fi
460
461
462
463	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
464	# AUFTEILEN
465	done=false
466	while [[ $done = false ]]
467	do
468	if (( $cputime <= 0 ))
469	then
470	printf "\n +++ wrong cpu-time or cpu-time missing"
471	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
472	printf "\n >>> "
473	read cputime 1>/dev/null 2>&1
474	else
475	done=true
476	fi
477	done
478	if [[ $remote_host = nech ]]
479	then
480	if (( tasks_per_node != 0 ))
481	then
482	(( cputime = cputime * tasks_per_node ))
483	elif [[ $numprocs != 0 ]]
484	then
485	(( cputime = cputime * numprocs ))
486	fi
487	fi
488	(( stunden = cputime / 3600 ))
489	(( resttime = cputime - stunden * 3600 ))
490	(( minuten = resttime / 60 ))
491	(( sekunden = resttime - minuten * 60 ))
492	timestring=${stunden}:${minuten}:${sekunden}
493
494
495
496	# PRUEFEN DER KERNSPEICHERANFORDERUNG
497	done=false
498	while [[ $done = false ]]
499	do
500	if (( memory <= 0 ))
501	then
502	printf "\n +++ wrong memory demand or memory demand missing"
503	printf "\n >>> Please type memory in MByte per process as INTEGER:"
504	printf "\n >>> "
505	read memory 1>/dev/null 2>&1
506	else
507	done=true
508	fi
509	done
510
511	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
512	then
513	if (( tasks_per_node != 0 ))
514	then
515	(( Memory = memory * tasks_per_node / 1000 ))
516	elif [[ $numprocs != 0 ]]
517	then
518	(( Memory = memory * numprocs / 1000 ))
519	else
520	(( Memory = memory / 1000 ))
521	fi
522	elif [[ $remote_host = lctit ]]
523	then
524	(( Memory = memory * tasks_per_node / 1000 ))
525	fi
526
527
528	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
529	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
530	then
531	(( memory = memory * threads_per_task ))
532	fi
533
534
535	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
536	if (( tasks_per_node != 0 ))
537	then
538	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
539	fi
540
541
542	# Calculate number of processes per node
543	(( processes_per_node = tasks_per_node * threads_per_task ))
544
545
546	# Calculate number of MPI tasks
547	(( mpi_tasks = numprocs / threads_per_task ))
548
549
550	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
551	if [[ "$scp_port" != "" ]]
552	then
553	PORTOPT="-P $scp_port"
554	SSH_PORTOPT="-p $scp_port"
555	fi
556
557
558	# HEADER-AUSGABE
559	if [[ $verify = true ]]
560	then
561	printf "\n\n"
562	printf "#--------------------------------------------------------------# \n"
563	spalte1=SUBJOB;spalte2=$(date)
564	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
565	printf "\| \| \n"
566	printf "\| values of parameters/options: \| \n"
567	spalte1=$(echo local_host$punkte \| cut -c-20)
568	spalte2=$punkte$local_host
569	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
570	spalte1=$(echo remote_host$punkte \| cut -c-20)
571	spalte2=$punkte$remote_host
572	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
573	spalte1=$(echo queue$punkte \| cut -c-20)
574	spalte2=$punkte$queue
575	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
576	spalte1=$(echo memory$punkte \| cut -c-20)
577	spalte2="$punkte$memory mb"
578	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
579	spalte1=$(echo cputime$punkte \| cut -c-20)
580	spalte2="$punkte$cputime sec"
581	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
582	spalte1=$(echo job_name$punkte \| cut -c-20)
583	spalte2="$punkte$job_name"
584	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
585	printf "#--------------------------------------------------------------# \n\n"
586
587
588	# KONTROLLABFRAGE, OB ALLES O.K.
589	antwort="dummy"
590	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
591	do
592	read antwort?" >>> continue (y/n) ? "
593	done
594	if [[ $antwort = n \|\| $antwort = N ]]
595	then
596	locat=verify; exit
597	fi
598	printf "\n"
599	fi
600
601
602
603	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
604	kennung=$RANDOM
605	job_on_remhost=${job_name}_${kennung}_$local_host
606	job_to_send=job_to_send_$kennung
607	if [[ $delete_dayfile = false ]]
608	then
609	remote_dayfile=${local_host}_${job_name}_result_$kennung
610	local_dayfile=${remote_host}_${job_name}
611	else
612	remote_dayfile=/dev/null
613	fi
614
615
616
617	# Generate the batch job scripts (qsub/msub/LoadLeveler)
618	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
619	then
620
621	# General LoadLeveler settings
622	execute_in_shell="#!/bin/ksh"
623	use_shell="# @ shell = /bin/ksh"
624	consumable_memory="ConsumableMemory($memory mb)"
625	class="# @ class = $queue"
626	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
627	network_to_use="# @ network.mpi = sn_all,shared,us"
628	data_limit="# @ data_limit = 1.76gb"
629	image_size="# @ image_size = 50"
630	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
631
632	if [[ $email_notification = none ]]
633	then
634	notify_user=""
635	else
636	notify_user="# @ notify_user = $email_notification"
637	if [[ $delete_dayfile = true ]]
638	then
639	notification='# @ notification = never'
640	fi
641	fi
642
643	if [[ $remote_host = ibmh ]]
644	then
645	data_limit=""
646	network_to_use=""
647	class="# @ class = $queue"
648	environment=""
649	rset="# @ rset = RSET_MCM_AFFINITY"
650	task_affinity="# @ task_affinity = core(1)"
651	elif [[ $remote_host = ibmkisti ]]
652	then
653	network_to_use="# @ network.MPI = sn_all,shared,US"
654	wall_clock_limit="# @ wall_clock_limit = $timestring"
655	if [[ $threads_per_task = 1 ]]
656	then
657	rset="# @ rset = RSET_MCM_AFFINITY"
658	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
659	fi
660	environment=""
661	use_shell=""
662	data_limit=""
663	image_size=""
664	elif [[ $remote_host = ibmku ]]
665	then
666	execute_in_shell="#!/usr/bin/ksh"
667	use_shell="# @ shell = /usr/bin/ksh"
668	consumable_memory=""
669	environment=""
670	network_to_use="# @ network.mpi = sn_all,shared,us"
671	data_limit=""
672	image_size=""
673	elif [[ $remote_host = ibms ]]
674	then
675	network_to_use="# @ network.mpi = csss,shared,us"
676	elif [[ $remote_host = ibmy ]]
677	then
678	consumable_memory=""
679	network_to_use=""
680	fi
681
682	cat > $job_to_send << %%END%%
683	$execute_in_shell
684	$use_shell
685
686	# @ job_type = parallel
687	# @ job_name = $job_name
688	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
689	# @ output = $remote_dayfile
690	# @ error = $remote_dayfile
691	$wall_clock_limit
692	$image_size
693	$class
694	$environment
695	$network_to_use
696	$data_limit
697	$rset
698	$mcm_affinity_options
699	$task_affinity
700	$notification
701	$notify_user
702
703	%%END%%
704
705	if (( nodes > 0 ))
706	then
707
708	if [[ $remote_host != ibmkisti ]]
709	then
710
711	cat >> $job_to_send << %%END%%
712	# @ node = $nodes
713	# @ tasks_per_node = $processes_per_node
714	# @ node_usage = $node_usage
715	# @ queue
716
717	%%END%%
718
719	else
720
721	cat >> $job_to_send << %%END%%
722	# @ total_tasks = $mpi_tasks
723	# @ blocking = unlimited
724	# @ queue
725
726	%%END%%
727
728	fi
729
730	else
731
732	if [[ $remote_host != ibmy ]]
733	then
734
735	cat >> $job_to_send << %%END%%
736	# @ blocking = unlimited
737	# @ total_tasks = $numprocs
738	# @ node_usage = $node_usage
739	# @ queue
740
741	%%END%%
742
743	else
744
745	cat >> $job_to_send << %%END%%
746	# @ node = 1
747	# @ total_tasks = $numprocs
748	# @ queue
749
750	%%END%%
751
752	fi
753
754	fi
755
756	# workaround because of silly job filter on ibmkisti
757	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
758	then
759	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
760	fi
761
762	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
763	then
764
765	cat > $job_to_send << %%END%%
766	#!/bin/ksh
767
768	# @ job_type = serial
769	# @ node_usage = $node_usage
770	# @ job_name = palm
771	# @ wall_clock_limit = ${timestring},$timestring
772	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
773	# @ output = $remote_dayfile
774	# @ error = $remote_dayfile
775	$class
776	$notification
777
778	# @ queue
779
780	%%END%%
781
782	elif [[ $remote_host = lccrayb ]]
783	then
784
785	if [[ $numprocs != 0 ]]
786	then
787	cat > $job_to_send << %%END%%
788	#!/bin/bash
789	#PBS -N $job_name
790	#PBS -l walltime=$timestring
791	#PBS -l mppwidth=$numprocs
792	#PBS -l mppnppn=$processes_per_node
793	#PBS -o $remote_dayfile
794	#PBS -j oe
795	#PBS -q $queue
796
797	$module_calls
798
799	%%END%%
800
801	else
802
803	continue
804
805	fi
806
807	elif [[ $remote_host = lccrayf ]]
808	then
809
810	if [[ $numprocs != 0 ]]
811	then
812	cat > $job_to_send << %%END%%
813	#!/bin/bash -l
814	#SBATCH -J $job_name
815	#SBATCH -t $timestring
816	#SBATCH -N $nodes
817	#SBATCH --ntasks-per-node=$processes_per_node
818	#SBATCH -p $queue
819	#SBATCH -o $remote_dayfile
820	#SBATCH -e $remote_dayfile
821
822	$init_cmds
823	$module_calls
824
825	%%END%%
826
827	else
828	cat > $job_to_send << %%END%%
829	#!/bin/bash -l
830	#SBATCH -J $job_name
831	#SBATCH -t $timestring
832	#SBATCH -l ncpus=1
833	#SBATCH -l pmem=${memory}mb
834	#SBATCH -m abe
835	#SBATCH -o $remote_dayfile
836	#SBATCH -e $remote_dayfile
837
838	$init_cmds
839	$module_calls
840
841	%%END%%
842
843	fi
844
845	elif [[ $remote_host = lcfimm ]]
846	then
847
848	if [[ $numprocs != 0 ]]
849	then
850	cat > $job_to_send << %%END%%
851	#!/bin/ksh
852	#PBS -N $job_name
853	#PBS -A $project_account
854	#PBS -l walltime=$timestring
855	#PBS -l nodes=${nodes}:ppn=$processes_per_node
856	#PBS -l pmem=${memory}mb
857	#PBS -m abe
858	#PBS -o $remote_dayfile
859	#PBS -j oe
860	mpd &
861
862	%%END%%
863
864	else
865	cat > $job_to_send << %%END%%
866	#!/bin/ksh
867	#PBS -N $job_name
868	#PBS -A $project_account
869	#PBS -l walltime=$timestring
870	#PBS -l ncpus=1
871	#PBS -l pmem=${memory}mb
872	#PBS -m abe
873	#PBS -o $remote_dayfile
874	#PBS -j oe
875
876	%%END%%
877
878	fi
879
880	elif [[ $remote_host = lcflow ]]
881	then
882	if [ $memory -gt 1800 ]; then
883	use_himem=""
884	else
885	use_himem="#"
886	fi
887	if [[ $numprocs != 0 ]]
888	then
889	pe_set="#$ -pe impi41 $numprocs"
890	else
891	pe_set="#$ -pe impi41 1"
892	fi
893	if [[ $queue = default ]]
894	then
895	queue_set=""
896	else
897	queue_set="#$ -q $queue"
898	fi
899	[[ "$disc_space" = "" ]] && disc_space=50
900
901	cat > $job_to_send << %%END%%
902	#!/bin/bash
903	#$ -S /bin/bash
904	#$ -N $job_name
905	#$ -cwd
906	#$ -l h_rt=$timestring
907	#$ -l h_vmem=${memory}M
908	#$ -o $remote_dayfile
909	#$ -j y
910	$pe_set
911	#$ -R y
912	${use_himem}#$ -l highmem=true
913	#$ -l h_fsize=${disc_space}G
914	$queue_set
915
916	%%END%%
917
918	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
919	then
920
921	if [[ $numprocs != 0 ]]
922	then
923	cat > $job_to_send << %%END%%
924	#!/bin/ksh
925	#PBS -N $job_name
926	#PBS -l walltime=$timestring
927	#PBS -l ncpus=$numprocs
928	#PBS -l pmem=${memory}mb
929	#PBS -o $remote_dayfile
930	#PBS -l nodes=$nodes:ppn=${processes_per_node}
931	#PBS -j oe
932
933	mpd &
934
935	%%END%%
936
937	else
938	cat > $job_to_send << %%END%%
939	#!/bin/ksh
940	#PBS -N $job_name
941	#PBS -l walltime=$timestring
942	#PBS -l ncpus=1
943	#PBS -l pmem=${memory}mb
944	#PBS -o $remote_dayfile
945	#PBS -j oe
946
947	%%END%%
948
949	fi
950
951	elif [[ $remote_host = lckiaps ]]
952	then
953
954	if [[ $numprocs != 0 ]]
955	then
956	cat > $job_to_send << %%END%%
957	#!/bin/ksh
958	#PBS -N $job_name
959	#PBS -l walltime=$timestring
960	#PBS -l select=1:ncpus=$numprocs
961	#PBS -l pmem=${memory}mb
962	#PBS -q $queue
963	#PBS -o $remote_dayfile
964	#PBS -j oe
965	#PBS -V
966
967	%%END%%
968
969	else
970	cat > $job_to_send << %%END%%
971	#!/bin/ksh
972	#PBS -N $job_name
973	#PBS -l walltime=$timestring
974	#PBS -l ncpus=1
975	#PBS -l pmem=${memory}mb
976	#PBS -o $remote_dayfile
977	#PBS -j oe
978
979	%%END%%
980
981	fi
982
983	elif [[ $remote_host = lcyon ]]
984	then
985
986	if [[ $numprocs != 0 ]]
987	then
988	cat > $job_to_send << %%END%%
989	#!/bin/ksh
990	#PBS -N $job_name
991	#PBS -l walltime=$timestring
992	#PBS -l ncpus=$numprocs
993	#PBS -l pmem=${memory}mb
994	#PBS -o $remote_dayfile
995	#PBS -j oe
996
997	%%END%%
998
999	else
1000	cat > $job_to_send << %%END%%
1001	#!/bin/ksh
1002	#PBS -N $job_name
1003	#PBS -l walltime=$timestring
1004	#PBS -l ncpus=1
1005	#PBS -l pmem=${memory}mb
1006	#PBS -o $remote_dayfile
1007	#PBS -j oe
1008
1009	%%END%%
1010
1011	fi
1012
1013	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1014	then
1015
1016	if [[ "$sgi_feature" != "" ]]
1017	then
1018	feature_directive="#PBS -l feature=$sgi_feature"
1019	else
1020	feature_directive=""
1021	fi
1022
1023	if [[ $queue = dataq \|\| $queue = permq ]]
1024	then
1025	feature_directive="#PBS -l feature=data"
1026	fi
1027
1028	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
1029	then
1030	queue_directive="#PBS -q $queue"
1031	else
1032	queue_directive=""
1033	fi
1034
1035	if [[ $email_notification = none ]]
1036	then
1037	email_directive=""
1038	else
1039	email_directive="#PBS -M $email_notification"
1040	fi
1041
1042	if [[ $numprocs != 0 ]]
1043	then
1044	cat > $job_to_send << %%END%%
1045	#!/bin/bash
1046	#PBS -N $job_name
1047	#PBS -l walltime=$timestring
1048	#PBS -l nodes=$nodes:ppn=${processes_per_node}
1049	#PBS -l naccesspolicy=$node_usage
1050	#PBS -o $remote_dayfile
1051	#PBS -j oe
1052	$feature_directive
1053	$queue_directive
1054	$email_directive
1055
1056	eval \`/sw/swdist/bin/modulesinit\`
1057	#. /usr/share/modules/init/bash
1058	$init_cmds
1059	$module_calls
1060
1061	echo ld_library_path=\$LD_LIBRARY_PATH
1062
1063	%%END%%
1064
1065	else
1066	cat > $job_to_send << %%END%%
1067	#PBS -S /bin/bash
1068	#PBS -N $job_name
1069	#PBS -l walltime=$timestring
1070	#PBS -l ncpus=1
1071	#PBS -o $remote_dayfile
1072	#PBS -j oe
1073	$queue_directive
1074	$email_directive
1075
1076	eval \`/sw/swdist/bin/modulesinit\`
1077	#. /usr/share/modules/init/bash
1078	$init_cmds
1079	$module_calls
1080
1081	%%END%%
1082
1083	fi
1084
1085	elif [[ $remote_host = lcxe6 ]]
1086	then
1087
1088	if [[ $numprocs != 0 ]]
1089	then
1090	cat > $job_to_send << %%END%%
1091	#!/bin/ksh
1092	#PBS -S /bin/ksh
1093	#PBS -N $job_name
1094	#PBS -A $project_account
1095	#PBS -j oe
1096	#PBS -l walltime=$timestring
1097	#PBS -l mppwidth=${numprocs}
1098	#PBS -l mppnppn=${processes_per_node}
1099	#PBS -m abe
1100	#PBS -o $remote_dayfile
1101	$email_directive
1102
1103	$init_cmds
1104	$module_calls
1105
1106	%%END%%
1107
1108	else
1109	cat > $job_to_send << %%END%%
1110	#!/bin/ksh
1111	#PBS -S /bin/ksh
1112	#PBS -N $job_name
1113	#PBS -A $project_account
1114	#PBS -j oe
1115	#PBS -l walltime=$timestring
1116	#PBS -l ncpus=1
1117	#PBS -l pmem=${memory}mb
1118	#PBS -m abe
1119	$email_directive
1120	#PBS -o $remote_dayfile
1121
1122	$init_cmds
1123	$module_calls
1124
1125	%%END%%
1126
1127	fi
1128
1129	elif [[ $remote_host = lckyoto ]]
1130	then
1131
1132	cat > $job_to_send << %%END%%
1133	#!/bin/ksh
1134	# @\$-o $remote_dayfile
1135	# @\$-eo -oi
1136	# @\$-lP 16
1137	# @\$-lp 1
1138	# @\$-lm 28gb -llm unlimited -ls unlimited
1139	# @\$-q $queue
1140	# @\$-Pvn abs_pack
1141	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1142	#. /thin/local/etc/setprofile/intel-11.0.sh
1143	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1144	. ~/.myprofile
1145	#. /home2/t/t51254/palm/current_version/myprofile
1146	#. /thin/apps/pgi/mpi.sh
1147	#
1148	env
1149	#
1150	set -x
1151
1152	%%END%%
1153
1154	elif [[ $remote_host = lcxt5m ]]
1155	then
1156
1157	if [[ $numprocs != 0 ]]
1158	then
1159	cat > $job_to_send << %%END%%
1160	#!/bin/ksh
1161	#PBS -S /bin/ksh
1162	#PBS -N $job_name
1163	#PBS -j oe
1164	#PBS -l walltime=$timestring
1165	#PBS -l mppwidth=${numprocs}
1166	#PBS -l mppnppn=${processes_per_node}
1167	#PBS -m abe
1168	#PBS -o $remote_dayfile
1169
1170	$init_cmds
1171	$module_calls
1172
1173	%%END%%
1174
1175	else
1176	cat > $job_to_send << %%END%%
1177	#!/bin/ksh
1178	#PBS -S /bin/ksh
1179	#PBS -N $job_name
1180	#PBS -j oe
1181	#PBS -l walltime=$timestring
1182	#PBS -l ncpus=1
1183	#PBS -l pmem=${memory}mb
1184	#PBS -m abe
1185	#PBS -o $remote_dayfile
1186
1187	$init_cmds
1188	$module_calls
1189
1190	%%END%%
1191
1192	fi
1193
1194	elif [[ $remote_host = lckyuh ]]
1195	then
1196	cat > $job_to_send << %%END%%
1197	#!/bin/bash
1198	#PJM -L "rscgrp=$queue"
1199	#PJM -L "node=$nodes"
1200	#PJM --mpi "proc=$numprocs"
1201	#PJM -L "elapse=$timestring"
1202	#PJM -o $remote_dayfile
1203	#PJM -j
1204	#PJM -X
1205	#PJM --no-stging
1206
1207	export LANG=en_US.UTF-8
1208	%%END%%
1209
1210	elif [[ $remote_host = lckyut ]]
1211	then
1212	cat > $job_to_send << %%END%%
1213	#!/bin/bash
1214	#PJM -L "rscgrp=$queue"
1215	#PJM -L "vnode=$numprocs"
1216	#PJM -L "vnode-core=1"
1217	#PJM -L "elapse=$timestring"
1218	#PJM --mpi proc=$numprocs
1219	#PJM -o $remote_dayfile
1220	#PJM -j
1221	#PJM -X
1222	#PJM --no-stging
1223
1224	export LANG=en_US.UTF-8
1225	%%END%%
1226
1227	elif [[ $remote_host = nech ]]
1228	then
1229
1230	if (( nodes > 1 ))
1231	then
1232	cat > $job_to_send << %%END%%
1233	#!/bin/ksh
1234	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1235	#PBS -l ${qsubmem}=${Memory}gb
1236	#PBS -b $nodes
1237	#PBS -o $remote_dayfile
1238	#PBS -N palm
1239	#PBS -j o
1240	#PBS -T mpisx
1241
1242	%%END%%
1243
1244	elif [[ $numprocs != 0 ]]
1245	then
1246	cat > $job_to_send << %%END%%
1247	#!/bin/ksh
1248	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1249	#PBS -l ${qsubmem}=${Memory}gb
1250	#PBS -o $remote_dayfile
1251	#PBS -N palm
1252	#PBS -j o
1253
1254	%%END%%
1255
1256	else
1257	cat > $job_to_send << %%END%%
1258	#!/bin/ksh
1259	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1260	#PBS -o $remote_dayfile
1261	#PBS -j o
1262
1263	%%END%%
1264
1265	fi
1266
1267	elif [[ $remote_host = necriam ]]
1268	then
1269
1270	if (( nodes > 1 ))
1271	then
1272	cat > $job_to_send << %%END%%
1273	#!/bin/ksh
1274	#PBS -b $nodes
1275	#PBS -o $remote_dayfile
1276	#PBS -N $job_name
1277	#PBS -j o
1278	#PBS -v MPIPROGINV=YES
1279
1280	%%END%%
1281
1282	elif [[ $numprocs != 0 ]]
1283	then
1284	cat > $job_to_send << %%END%%
1285	#!/bin/ksh
1286	#PBS -q ${queue}
1287	#PBS -o $remote_dayfile
1288	#PBS -N $job_name
1289	#PBS -j o
1290	#PBS -v MPIPROGINV=YES
1291
1292	%%END%%
1293
1294	else
1295	cat > $job_to_send << %%END%%
1296	#!/bin/ksh
1297	#PBS -o $remote_dayfile
1298	#PBS -j o
1299
1300	%%END%%
1301
1302	fi
1303
1304	elif [[ $remote_host = lctit ]]
1305	then
1306	cat > $job_to_send << %%END%%
1307	#!/bin/ksh
1308	$init_cmds
1309	$module_calls
1310
1311	%%END%%
1312
1313	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1314	if [[ $tasks_per_node != $processes_per_node ]]
1315	then
1316	submcom="$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 $queue "
1317	else
1318	submcom="$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 "
1319	fi
1320
1321	else
1322
1323	cat > $job_to_send << %%END%%
1324	# @\$-q ${queue}
1325	# @\$-l${qsubtime} $timestring
1326	# @\$-l${qsubmem} ${memory}mb
1327	# @\$-o $remote_dayfile
1328	# @\$-eo
1329
1330	%%END%%
1331
1332	fi
1333
1334
1335	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1336	# BEI EXIT VERANLASSEN
1337	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1338	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1339	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1340	then
1341	echo "set +vx" >> $job_to_send
1342	echo "trap '" >> $job_to_send
1343	echo "set +vx" >> $job_to_send
1344	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* ]]
1345	then
1346	if [[ $remote_host = ibmh ]]
1347	then
1348	return_queue=c1
1349	elif [[ $remote_host = ibmkisti ]]
1350	then
1351	return_queue=class.1-2
1352	elif [[ $remote_host = ibmku ]]
1353	then
1354	return_queue=sdbg2
1355	elif [[ $remote_host = ibms ]]
1356	then
1357	return_queue=p_normal
1358	elif [[ $remote_host = ibmy ]]
1359	then
1360	return_queue=serial
1361	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1362	then
1363	return_queue=serialq
1364	elif [[ $remote_host = necriam ]]
1365	then
1366	return_queue=SP
1367	elif [[ $remote_host = lccrayb ]]
1368	then
1369	return_queue=dataq
1370	elif [[ $remote_host = lckiaps ]]
1371	then
1372	return_queue=express
1373	elif [[ $remote_host = lckyuh ]]
1374	then
1375	return_queue=cx-single
1376	elif [[ $remote_host = lckyut ]]
1377	then
1378	return_queue=cx-single
1379	else
1380	return_queue=unknown
1381	fi
1382
1383	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1384	then
1385
1386	if [[ $remote_host = ibmku ]]
1387	then
1388	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1389	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1390	else
1391	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1392	fi
1393	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1394	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1395	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1396	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1397	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1398	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1399	if [[ $host != "ibmh" ]]
1400	then
1401	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1402	fi
1403	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1404	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1405
1406	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1407	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1408
1409	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1410	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
1411	if [[ $remote_host = ibmku ]]
1412	then
1413	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1414	fi
1415	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1416
1417	elif [[ $remote_host = nech ]]
1418	then
1419	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1420	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1421	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1422	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1423	echo "#PBS -j o" >> $job_to_send
1424	echo " " >> $job_to_send
1425	echo "set -x" >> $job_to_send
1426	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1427	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1428	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1429	echo "%%END%%" >> $job_to_send
1430
1431	elif [[ $remote_host = necriam ]]
1432	then
1433	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1434	echo "#PBS -q $return_queue" >> $job_to_send
1435	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1436	echo "#PBS -j o" >> $job_to_send
1437	echo " " >> $job_to_send
1438	echo "set -x" >> $job_to_send
1439	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1440	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1441	echo "%%END%%" >> $job_to_send
1442
1443	elif [[ $remote_host = lckyuh ]]
1444	then
1445	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1446	echo "#!/bin/bash" >> $job_to_send
1447	echo "#PJM -L \"node=1\"" >> $job_to_send
1448	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1449	echo "#PJM --no-stging" >> $job_to_send
1450	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1451	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1452	echo "#PJM -j" >> $job_to_send
1453	echo " " >> $job_to_send
1454	echo "export LANG=en_US.UTF-8" >> $job_to_send
1455	echo "set -x" >> $job_to_send
1456	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1457	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1458	echo "%%END%%" >> $job_to_send
1459
1460	elif [[ $remote_host = lckyut ]]
1461	then
1462	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1463	echo "#!/bin/bash" >> $job_to_send
1464	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1465	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1466	echo "#PJM --no-stging" >> $job_to_send
1467	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1468	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1469	echo "#PJM -j" >> $job_to_send
1470	echo " " >> $job_to_send
1471	echo "export LANG=en_US.UTF-8" >> $job_to_send
1472	echo "set -x" >> $job_to_send
1473	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1474	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1475	echo "%%END%%" >> $job_to_send
1476
1477	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1478	then
1479	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1480	echo "#!/bin/bash" >> $job_to_send
1481	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1482	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1483	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1484	echo "#PBS -l feature=data" >> $job_to_send
1485	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1486	echo "#PBS -j oe" >> $job_to_send
1487	echo " " >> $job_to_send
1488	echo ". /usr/share/modules/init/bash" >> $job_to_send
1489	echo "set -x" >> $job_to_send
1490	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1491	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1492	echo "%%END%%" >> $job_to_send
1493
1494	elif [[ $remote_host = lccrayb ]]
1495	then
1496	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1497	echo "#!/bin/bash" >> $job_to_send
1498	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1499	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1500	echo "#PBS -l mppwidth=1" >> $job_to_send
1501	echo "#PBS -l mppnppn=1" >> $job_to_send
1502	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1503	echo "#PBS -j oe" >> $job_to_send
1504	echo " " >> $job_to_send
1505	echo "set -x" >> $job_to_send
1506	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1507	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1508	echo "%%END%%" >> $job_to_send
1509
1510	elif [[ $remote_host = lcflow ]]
1511	then
1512	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1513	echo "#!/bin/bash" >> $job_to_send
1514	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1515	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1516	echo "SGEPREFIX -cwd" >> $job_to_send
1517	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1518	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1519	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1520	echo "SGEPREFIX -j y" >> $job_to_send
1521	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1522	echo " " >> $job_to_send
1523	echo "set -x" >> $job_to_send
1524	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1525	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1526	echo "" >> $job_to_send
1527	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1528	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1529	echo "rm -f scpjob.${kennung}" >> $job_to_send
1530	echo "%%END%%" >> $job_to_send
1531	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1532	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1533
1534	else
1535
1536	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1537	echo "# @\\\$-q $return_queue" >> $job_to_send
1538	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1539	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1540	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1541	then
1542	echo "# @\$-l mpp_p=0" >> $job_to_send
1543	fi
1544	echo '# @\$-lF 10mb' >> $job_to_send
1545	# echo '# @\$-o /dev/null' >> $job_to_send
1546	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1547	echo '# @\\\$-eo' >> $job_to_send
1548	echo " " >> $job_to_send
1549	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1550	then
1551	echo "set +vx" >> $job_to_send
1552	echo ". .profile" >> $job_to_send
1553	fi
1554	echo "set -x" >> $job_to_send
1555	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
1556	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1557	echo "%%END%%" >> $job_to_send
1558	fi
1559
1560	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1561	then
1562	echo "llsubmit scpjob.$kennung" >> $job_to_send
1563	elif [[ $remote_host = lccrayb ]]
1564	then
1565	echo "qsub -q $return_queue scpjob.$kennung" >> $job_to_send
1566	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1567	then
1568	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1569	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1570	echo "msub scpjob.$kennung" >> $job_to_send
1571	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1572	then
1573	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1574	elif [[ $remote_host = t3es ]]
1575	then
1576	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1577	elif [[ $remote_host = lckiaps ]]
1578	then
1579	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1580	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1581	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1582	elif [[ $remote_host = lckyu* ]]
1583	then
1584	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1585	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1586	elif [[ $remote_host = lcflow ]]
1587	then
1588	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1589	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1590	else
1591	echo "$submcom scpjob.$kennung" >> $job_to_send
1592	fi
1593	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1594	then
1595	echo "rm scpjob.$kennung" >> $job_to_send
1596	fi
1597	if [[ $remote_host = nech ]]
1598	then
1599	echo "cd -" >> $job_to_send
1600	fi
1601	else
1602	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1603	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1604	fi
1605	echo "set -x" >> $job_to_send
1606	echo " ' exit" >> $job_to_send
1607	echo "set -x" >> $job_to_send
1608	fi
1609
1610
1611
1612	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1613	cat $file_to_send >> $job_to_send
1614
1615	if [[ $remote_host = ibm ]]
1616	then
1617	echo " " >> $job_to_send
1618	echo "exit" >> $job_to_send
1619	fi
1620
1621	# remove job file
1622	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1623	then
1624	echo " " >> $job_to_send
1625	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1626	fi
1627
1628
1629
1630	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1631	if [[ -z $remote_user ]]
1632	then
1633	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1634	then
1635	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1636	fi
1637	fi
1638
1639
1640
1641	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1642	if [[ $no_submit = false ]]
1643	then
1644	if [[ $remote_host != $local_host ]]
1645	then
1646	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1647	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1648	then
1649	job_catalog_save=$job_catalog
1650	job_catalog=job_queue
1651	elif [[ $remote_host = nech ]]
1652	then
1653	job_catalog_save=$job_catalog
1654	job_catalog=/hpf/b/${remote_user}/job_queue
1655	fi
1656	if [[ $remote_host = nech ]]
1657	then
1658	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1659	# TRANSFERIERT WERDEN
1660	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1661	else
1662	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1663	fi
1664	if [[ $? = 1 ]]
1665	then
1666	locat=scp; exit
1667	fi
1668	if [[ $remote_host = ibms ]]
1669	then
1670	job_catalog=$job_catalog_save
1671	fi
1672	[[ $verify = true ]] && printf "\n >>> finished\n"
1673	else
1674	eval job_catalog=$job_catalog
1675	cp $job_to_send ${job_catalog}/$job_on_remhost
1676	fi
1677
1678
1679
1680	# NQS- BZW. LOADLEVELER-JOB STARTEN
1681	if [[ $remote_host != $local_host ]]
1682	then
1683	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1684
1685	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1686	then
1687	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1688	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1689	elif [[ $remote_host = ibmku ]]
1690	then
1691	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1692	elif [[ $remote_host = lcflow ]]
1693	then
1694	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1695	else
1696	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1697	fi
1698
1699	[[ $verify = true ]] && printf " >>> o.k.\n"
1700	else
1701	cd $job_catalog
1702	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1703	then
1704	eval $submcom $job_on_remhost
1705	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1706	then
1707	chmod u+x $job_on_remhost
1708	eval $submcom $job_on_remhost
1709	elif [[ $local_host = nech ]]
1710	then
1711	if [[ $queue = default ]]
1712	then
1713	eval $submcom $job_on_remhost
1714	else
1715	eval $submcom -q $queue $job_on_remhost
1716	fi
1717	else
1718	qsub $job_on_remhost
1719	fi
1720
1721	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1722	# only at the end of the job.
1723	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1724	then
1725	rm $job_on_remhost
1726	fi
1727	cd - > /dev/null
1728	fi
1729	fi
1730
1731
1732
1733	# ABSCHLUSSARBEITEN
1734
1735	if [[ $no_submit = false ]]
1736	then
1737	rm -f $job_to_send
1738	fi
1739	[[ $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 |